CINXE.COM
The National Academies Press
<!doctype html> <html lang="en"> <head> <meta charset="UTF-8"> <meta http-equiv="X-UA-Compatible" content="IE=edge"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <meta http-equiv="cache-control" content="no-cache"/> <title>The National Academies Press</title> <script> // if (document.location.hostname.search(/nap\.edu$/) < 0) { // url = "http://www.na" + "p.edu" + document.location.pathname + document.location.search // document.location.replace(url); // } </script> <!--[if lt IE 9]> <script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script> <script src="/includes/js/svg4everybody.ie8.min.js"></script> <![endif]--> <meta name="alexaVerifyID" content="02J2SepmgvpkhVXcfeE98cX3HKs" /> <link href="//nap.nationalacademies.org/stylesheets/nasem-nap.css" rel="stylesheet" type="text/css"> <link href="//nap.nationalacademies.org/stylesheets/topnav-lite-min.css" rel="stylesheet" type="text/css" /> <link href="//nap.nationalacademies.org/stylesheets/component.css" rel="stylesheet" type="text/css" /> <!--[if gte IE 5]> <style type="text/css"> #blanket {filter:alpha(opacity=65);}</style><![endif]--> <script src="/includes/js/jquery-3.5.1.min.js"></script> <script src="/includes/js/jquery.dotdotdot.min.js" type="text/javascript"></script> <!--<script src="https://www.nationalacademies.org//_cache_4662/_themesdelivery/nasuwTheme/assets/vendor.js"></script>--> <!-- <script src="/includes/js/jquery.min-3.1.0.js"></script> --> <!-- MC Goal Tracking --> <script type="text/javascript"> var $mcGoal = {'settings':{'uuid':'eaea39b6442dc4e0d08e6aa4a','dc':'us4'}}; (function() { var sp = document.createElement('script'); sp.type = 'text/javascript'; sp.async = true; sp.defer = true; sp.src = ('https:' == document.location.protocol ? 'https://s3.amazonaws.com/downloads.mailchimp.com' : 'http://downloads.mailchimp.com') + '/js/goal.min.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(sp, s); })(); </script> <!-- Data Layer --> <script> window.dataLayer = window.dataLayer || [];dataLayer.push({ 'pageTitle' : "Home", 'loginStatus' : "guest", 'form' : "subscribe" }); </script> <script>(function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start': new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0], j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src= 'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f); })(window,document,'script','dataLayer','GTM-N9MJ23');</script> <!-- Global site tag (gtag.js) - Google AdWords: 1071160892 --> <script async src="https://www.googletagmanager.com/gtag/js?id=AW-1071160892"></script> <script> window.dataLayer = window.dataLayer || []; function gtag(){dataLayer.push(arguments);} gtag('js', new Date()); gtag('config', 'AW-1071160892'); gtag('config', 'AW-1071160892'); </script> <script id="mcjs">!function(c,h,i,m,p){m=c.createElement(h),p=c.getElementsByTagName(h)[0],m.async=1,m.src=i,p.parentNode.insertBefore(m,p)}(document,"script","https://chimpstatic.com/mcjs-connected/js/users/eaea39b6442dc4e0d08e6aa4a/208cb1a4b431cae38b54bdb6f.js");</script> <link rel="stylesheet" href="/stylesheets/vendor/bootstrap-3.0.3/bootstrap.min.css"> <link rel="stylesheet" href="/stylesheets/naphome.css" /> <style> body { /*background-color: #FAF8F6; */ background-color: #f0ede5; } span.searchterm { color: #333; } .sibcol2 { background-color: #fff !important; } .chaphits { background: #fff; padding: 10px 60px 10px 10px; margin-bottom: 10px; } .chaphitshdr { background: #dedede; padding: 10px 60px 10px 10px; margin-bottom: 10px; } .chaphits h4, .chaphitshdr h4 { line-height: 1.3 !important; color: #666; } .btn-hits { color: #999; background-color: #efefef !important; border: none; border-radius: 0px !important; } .print a.buyit { width: 100%; background-color: #1d1646; color: white; padding: 5px; } .print a.buyit:hover { background: #7899F2; color: white; } a.buyit { display: inline-block; padding: 6px 12px 6px 35px; background: none; margin: 10px 0px; color: #fff; } /*.btn-nohits a { cursor: default; } */ .format { text-transform: capitalize !important; } .webprice { font-size: 16px; } a.dloadit { white-space: nowrap; } /*.btn-nohits a:hover { text-decoration: none; } */ .heading.row p.small { font-size: 90%; color: #999; padding: 5px 0px 5px 0px; margin: 0px; } .form-group.searchinside { width: 90%; } .searchbook { display: none; border-top: 1px solid #ccc; padding: 10px 0px 10px 0px; background: #ccc; margin-bottom: 20px; } .sib_chapname { padding-top: 5px; font-size: 110%; } .sib_chapname_nohits, .btn-nohits { display: none; } .sib_chapname_nohits .viewlinks { display: none; } div.fromblock { line-height: 18pt; color: #000; font-style: italic; font-weight: bold; } .searchagain { font-size: 12px; } .chaphit_desc { border-top: 1px solid #ccc; border-bottom: 1px solid #ccc; padding-left: 0px; } .chaphit_row { margin-bottom: 20px; display: block; } .chaphit_row_nodisplay { margin: 0px; display: none; } .hitlink { font-size: 85%; } .mask { height: 0px; opacity: 0; -moz-transition: all .5s ease-in-out; -webkit-transition: all .5s ease-in-out; position: absolute; bottom: 0px; left: 0; width: 100%; background-color: #4678aa; color: white; padding-top: 10px; } .btn-hits:hover .mask { opacity: .8; height: 100%; bottom: 0px; color: white; } .siccol2 {} .row.inchap_results { background: #fff; margin-bottom: 10px; padding-top: 10px; padding-bottom: 10px; } /*Styles to display page location*/ .termincontext { color: #F67F0B; font-family: 'archivo_narrowbold', helvetica, arial; } .sic_paragraph { color: #666; } #openbook_buying_info p { font-size: 12px; } </style> <style>pre { background-color: transparent !important; }</style> </head> <body> <!-- Google Tag Manager (nos cript) --> <noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-N9MJ23" height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript> <!-- End Google Tag Manager (noscript) --> <a name="top" id="top"></a> <a href="#content" class="sr-only sr-only-focusable">Skip to main content</a> <style> body { overflow-x: hidden; } #lowerMenu { background: #fff } .global-nav--local .mega-menu { display: block; -webkit-transform: translateY(0em); transform: translateY(0em); } .global-nav--local.search-is-hidden .mega-menu { display: none; -webkit-transform: translateY(0em); transform: translateY(0em); } .global-nav .container:not(.mega-menu__header) { padding: 0rem; padding-top: 0rem; padding-right: 0rem; padding-top: 0.5rem; padding-right: 1rem; } .global-nav__controls { top: .3rem; padding-right: 20px; } .global-nav--local .global-nav__links { -webkit-transform: translateY(0); transform: translateY(0); max-width: 75%; float: right; background: #fff; } .global-nav__links ul { display: -webkit-flex; display: flex; -webkit-justify-content: flex-end; justify-content: flex-end; margin-top: .5rem; } .global-nav--local.menu-is-hidden .global-nav__links { opacity: 0; -webkit-transform: translateY(0.5em); transform: translateY(0.5em); visibility: hidden; } #header-large { max-width: 1180px; margin: 0; padding: 0 0; -webkit-box-shadow: none; -moz-box-shadow: none; box-shadow: none; flex-wrap: nowrap; position: absolute; top: 113px; text-align: right; width: 100%; right: 0; } #header-large .user-links { white-space: nowrap; display: inline-flex; margin: 0; padding: 0; background: #fff; z-index: 0; list-style: none; min-width:500px; } .user-links li { margin-right: 7%; } .user-links li a { font-size: 10px; text-align: left; text-transform: uppercase; color: #000; } #header-large a.site-links { margin-right: 15px; color: #000; padding-left: 0px; padding-right: 5px; font-size: 12px; text-transform: uppercase; } .site-links:hover { text-decoration: underline; } nav#sitelinks-wide { position: absolute; top: 16px; } #sitelinks-wide ul { display: flex; justify-content: space-between; margin: 0; padding: 0; } #sitelinks-wide ul li { #sitelinks-mobile { display: none; } white-space: nowrap; } #sitelinks-mobile { display: none; } .user-links li a:hover { text-decoration: underline; } .dropdown-menu { position: absolute; top: 100%; left: 10px; z-index: 9999999; display: none; float: left; min-width: 160px; padding: 5px 0; margin: 2px 0 0; list-style: none; font-size: 14px; background-color: #ffffff; border: 1px solid #cccccc; border: 1px solid rgba(0, 0, 0, 0.15); border-radius: 4px; -webkit-box-shadow: 0 6px 12px rgba(0, 0, 0, 0.175); box-shadow: 0 6px 12px rgba(0, 0, 0, 0.175); background-clip: padding-box; } #dropdownMenuButton { border: 1px solid #ccc; font-size: 14px; font-weight: 700; } .dropdown #dropdownMenuButton:after { content: '\02795'; /* Unicode character for "plus" sign (+) */ font-size: 13px; color: #777; float: right; margin-left: 5px; } .dropdown.open #dropdownMenuButton:after { content: "\2796"; /* Unicode character for "minus" sign (-) */ } li.naphelp { display: block;} @media (max-width: 930px) { #header-large { right: 80px; } #header-large .user-links { min-width:100px; } nav#sitelinks-wide { display: none; } #sitelinks-mobile { display: block; position: absolute; top: -10px; left: 107px; } #sitelinks-wide ul { display: block; } #sitelinks-wide ul li { margin-right: 0%; white-space: nowrap; } .dropdown-item { display: block; width: 100%; padding: .25rem 1.5rem; clear: both; font-weight: 400; color: #212529; text-align: inherit; white-space: nowrap; background-color: transparent; border: 0; text-align: left; } .dropdown.nap-mega-menu.mega-dropdown { display: none;} li.naphelp { display: none;} } .global-nav__links { display: none; opacity: 1; transition: opacity .15s, -webkit-transform .15s; transition: opacity .15s, transform .15s; transition: opacity .15s, transform .15s, -webkit-transform .15s; } .global-nav--local .global-nav__links { -webkit-transform: translateY(0); transform: translateY(0); } .global-nav--local.menu-is-hidden .global-nav__links { opacity: 0; -webkit-transform: translateY(0.5em); transform: translateY(0.5em); } .global-nav__links ul { display: -webkit-flex; display: flex; -webkit-justify-content: flex-end; justify-content: flex-end; margin-top: 1rem; } .global-nav__links li:hover .global-nav__link>span::before, .global-nav__links li:focus-within .global-nav__link>span::before, .global-nav__links .global-nav__link.is-active>span::before, .global-nav__links .global-nav__link:hover>span::before, .global-nav__links .global-nav__link:focus>span::before { background-color: #1C143D; ; } .global-nav__links li:hover .global-nav__link>span::before::after, .global-nav__links li:focus-within .global-nav__link>span::before::after, .global-nav__links .global-nav__link.is-active>span::before::after, .global-nav__links .global-nav__link:hover>span::before::after, .global-nav__links .global-nav__link:focus>span::before::after { opacity: 1; } @media screen and (min-width: 65em) { .global-nav__links { display: block; } .global-nav__links>ul>li:last-of-type .global-nav__link { margin-right: -1.25em; } } .global-nav__link { position: relative; display: block; padding: .75em 1.5em .5em; color: rgb(34, 34, 34); color: var(--color-neutral--charcoal); border-bottom: 5px solid transparent; /*font-size: .885rem;*/ font-weight: 700; font-weight: var(--lato--bold); text-decoration: none; transition: color .15s; } .global-nav__link>span { position: relative; } .global-nav__link>span::before { content: ""; position: absolute; bottom: -.5em; left: 0; width: 100%; height: 5px; background-color: transparent; transition: background-color .15s; } .global-nav__dropdown { position: absolute; top: calc(100% + 1em); left: 50%; display: none; width: 600px; background-color: rgb(255, 255, 255); background-color: var(--color-neutral--white); box-shadow: 0 5px 15px rgba(34, 34, 34, .25); box-shadow: 0 5px 15px var(--box-shadow--charcoal); -webkit-transform: translateX(-50%); transform: translateX(-50%); z-index: 20; } .global-nav__dropdown::before { content: ""; position: absolute; top: -1em; left: 0; display: block; width: 100%; height: 1em; } .global-nav__dropdown .dropdown__left { position: relative; min-height: 260px; width: 40%; padding: 1.5rem; background: rgb(255,255,255); background: -moz-linear-gradient(180deg, rgba(255,255,255,1) 0%, rgba(73,69,104,1) 0%, rgba(0,0,0,1) 100%, rgba(195,228,228,1) 100%); background: -webkit-linear-gradient(180deg, rgba(255,255,255,1) 0%, rgba(73,69,104,1) 0%, rgba(0,0,0,1) 100%, rgba(195,228,228,1) 100%); background: linear-gradient(180deg, rgba(255,255,255,1) 0%, rgba(73,69,104,1) 0%, rgba(0,0,0,1) 100%, rgba(195,228,228,1) 100%); filter: progid:DXImageTransform.Microsoft.gradient(startColorstr="#ffffff",endColorstr="#c3e4e4",GradientType=1); } .global-nav__dropdown .dropdown__left a { display: block; color: rgb(255, 255, 255); color: var(--color-neutral--white); font-weight: 400; font-weight: var(--lato--regular); transition: color .15s; font-size: 16px; } .global-nav__dropdown .dropdown__left a:hover, .global-nav__dropdown .dropdown__left a:focus, .global-nav__dropdown .dropdown__left a:active { color: rgb(246,210,0); color: var(--color-yellow--pure); } .global-nav__dropdown .dropdown__left ul { -webkit-flex-direction: column; flex-direction: column; } .global-nav__dropdown .dropdown__left li a { padding: .5em 0; line-height: 1.4; } .global-nav__dropdown .dropdown__left--default { position: absolute; top: 0; right: 0; bottom: 0; left: 0; overflow: hidden; background-color: rgb(243, 241, 235); background-color: var(--color-brown--bone); } .global-nav__dropdown .dropdown__left--default img { width: 100%; height: auto; /* stylelint-disable max-nesting-depth */ /* stylelint-enable max-nesting-depth */ } @supports (object-fit: cover) { .global-nav__dropdown .dropdown__left--default img { height: 100%; object-fit: cover; } } .global-nav__dropdown .dropdown__right { width: 60%; padding: 1.5rem; } .global-nav__dropdown .dropdown__right .button { margin-top: 2em; text-shadow: none; } .global-nav__dropdown .dropdown__right a { color: rgb(0, 89, 145); color: var(--color-blue--bahama); text-decoration: none; transition: color .15s; } .global-nav__dropdown .dropdown__right a:hover, .global-nav__dropdown .dropdown__right a:focus { color: rgb(33,23,71); color: var(--color-blue--imperial); text-decoration: underline; } .rm-dropdown:hover::after, .rm-dropdown:focus::after { visibility: hidden; } .has-dropdown { position: relative; } .has-dropdown::after { content: ""; position: absolute; top: calc(100% + 3px); left: 50%; width: 26px; height: 15px; background-image: url(data:image/png;base64,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); background-repeat: no-repeat; -webkit-transform: translateX(-50%); transform: translateX(-50%); visibility: hidden; } .has-dropdown:hover::after, .has-dropdown:focus::after { visibility: visible; } .has-dropdown:hover .global-nav__dropdown { display: -webkit-flex; display: flex; } .has-dropdown:focus-within .global-nav__dropdown { display: -webkit-flex; display: flex; } .has-dropdown:nth-last-of-type(-n+3) .global-nav__dropdown { left: auto; -webkit-transform: none; transform: none; } .has-dropdown:nth-last-of-type(3) .global-nav__dropdown { right: -200%; } .has-dropdown:nth-last-of-type(2) .global-nav__dropdown { right: -90%; } .has-dropdown:last-of-type .global-nav__dropdown { right: 5%; } .global-nav__links { z-index: 5; } header.global-nav { height: 150px; } .button.button--secondary.button--outline.button--icon { font-size: 12px; padding-right: 40px; } .button.button--icon svg { height: 2.5em; position: absolute; top: 50%; left: 60%; -webkit-transform: translate(-50%, -50%); transform: translate(-50%, -50%); width: 24px; } .controls__search, .controls__menu, .controls__nav { font-size: 12px; } .global-nav__menus--upper { display: -webkit-flex; display: flex; -webkit-justify-content: space-between; justify-content: space-between; margin-bottom: 2rem; } .global-nav__link { font-size: 16px; } .global-nav__links ul { display: -webkit-flex; display: flex; -webkit-justify-content: flex-end; justify-content: flex-end; margin-top: 5px; list-style: none; } .invert { background: #000; color: #fff; } .controls__nav .menu-off { display: inline; } .controls__nav .menu-on { display: none; } .controls__nav.invert .menu-off { display: none } .controls__nav.invert .menu-on { display: inline; } .mega-menu { background-color: rgb(243, 241, 235); background-color: var(--color-brown--bone); position: fixed; top: 0; right: 0; bottom: 0; left: 0; overflow: scroll; z-index: 9999999; } .mega-menu .mega-menu__header { display: -webkit-flex; display: flex; -webkit-justify-content: space-between; justify-content: space-between; -webkit-align-items: center; align-items: center; padding-top: 1rem; } .mega-menu__content .search-form__container { position: relative; max-width: 70em; margin: 0 auto; } #search .global-nav__logo { position: absolute; top: 1em; } .sf-top { margin-top: 50px; } #searchFormNAP { display: block; height: 100px; } a:hover, a:focus { color: #1C143D; text-decoration: underline; } #fixedfooter a:hover, #fixedfooter a:focus { color: #fff; text-decoration: underline; } .btn-blueflat.btn-searchhome.hide-xs { display: none; } .mega-menu__content { margin-top: 50px; margin-left: 7px; } .accordion__title.faq__accordion__title .flex.justify-between { display: inherit; } .global-nav__logo { margin: 0px !important; position: absolute; left: 15px; top: 25px; height: 55px; } .global-nav__link:hover { text-decoration: none; } @media screen and (min-width: 65em) { .mega-menu__content .search-form { padding: 1em 0px 3em 0em; } } @media (min-width: 45rem) { .global-nav__logo {} } @media (max-width: 480px) { .global-nav__logo { margin: 0px !important; position: absolute; left: 15px; top: 12px; } } /* Style the buttons that are used to open and close the accordion panel */ #mobile-menu button.accordion, #mobile-menu button.no-accordion { letter-spacing: 1px; text-align: left; color: #4148C2; font-size: 1.1em; font-weight: 700 !important; font-family: 'lato', helvetica, sans-serif; padding: 15px 0px 15px 0px; border-top: none !important; border-bottom: .5px solid rgb(221, 213, 199) !important; cursor: pointer; width: 100%; } .mega-menu__content .accordion { border: 0; font-weight: 700; } #mobile-menu .accordion:after { content: '\02795'; /* Unicode character for "plus" sign (+) */ font-size: 13px; color: #777; float: right; margin-left: 5px; } #mobile-menu button.active:after { content: "\2796"; /* Unicode character for "minus" sign (-) */ } /* Add a background color to the button if it is clicked on (add the .active class with JS), and when you move the mouse over it (hover) */ #mobile-menu button.active, #mobile-menu button.accordion:hover { background-color: transparent; } /* Style the accordion panel. Note: hidden by default */ #mobile-menu .panel { padding: 0 0; background-color: transparent; max-height: 0; box-shadow: 0 !important; overflow: hidden; transition: max-height 0.2s ease-out; margin: 0px !important; border: 0 !important; } #mobile-menu .panel ul li { margin: 0px 0px; padding-left: 0px; } #mobile-menu .panel ul { margin-top: 10px; margin-bottom: 10px; padding-left: 0px; } #mobile-menu .panel ul li { line-height: 2; } #mobile-menu .panel ul li a { color: #000; text-decoration: underline; cursor: pointer; } .controls__menu-mobile { text-transform: uppercase; } .controls__menu-mobile svg { fill: #222; fill: var(--color-neutral--charcoal); width: 1.75rem; height: 1.75rem; } .cart-badge { display: inline-block; text-align: center; line-height: 1; background: #7899F2; color: white; font-weight: bold; border-radius: 20px; padding: 3px 6px; margin-left: 3px; } @media screen and (min-width: 65em) { #mobile-menu .controls__menu { display: inline; } } @media screen and (max-width: 65em) { #mobile-menu .controls__menu svg { display: inline; } } /*NAP mega menu dropdown*/ li.nap-mega-menu.mega-dropdown { position: static !important; width:100% !important; } .nap-mega-menu .mega-dropdown-menu { padding: 20px 0px; box-shadow: none; -webkit-box-shadow: none; margin-top: 10px !important; box-shadow: 0 5px 15px var(--box-shadow--charcoal); transform: none !important; top: 100% !important; } .nap-mega-menu .mega-dropdown-menu > li > ul.browse-col { padding: 0; margin: 0; width: 100%; } .nap-mega-menu .mega-dropdown-menu > li > ul > li { list-style: none; } .nap-mega-menu .mega-dropdown-menu > li > ul > li > a { display: block; padding: 3px 20px; clear: both; font-weight: normal; line-height: 1.428571429; color: #4148C2; white-space: normal; font-size: 14px; text-transform: none; } .nap-mega-menu .mega-dropdown-menu > li ul > li > a:hover, .nap-mega-menu .mega-dropdown-menu > li ul > li > a:focus { text-decoration: none; color: #444; background-color: #f5f5f5; } .nap-mega-menu .mega-dropdown-menu .dropdown-header a { color: #4148C2; font-size: 14px; font-weight: 900 !important; text-align: left; text-transform: uppercase !important; padding: 0px; margin-bottom: 7px; } .nap-mega-menu .mega-dropdown-menu form { margin:3px 20px; } .nap-mega-menu .mega-dropdown-menu .form-group { margin-bottom: 3px; } .nap-mega-menu .vertical-divider { border-left: 1px solid #ccc; } .nap-mega-menu .mega-dropdown-menu > li.d-flex { display: inline-flex !important; margin-right: 0 !important; justify-content: space-between; flex-wrap: nowrap !important; flex-direction: row !important; } #nap-mega-menu-inner { justify-content: space-between; align-content:start; } #nap-mega-menu-inner li.browse-col { width: 22%; } </style> <header class="global-nav global-nav--local menu-is-hidden search-is-hidden" id="lowerMenu"> <div class="container flex justify-between"> <!--Mega menu start --> <div aria-hidden="true" class="mega-menu" id="search"> <div data-lock=""> <div class="container mega-menu__header"> <div class="global-nav__logo"> <a class="navbar-brand js-scroll-trigger" href="https://www.nationalacademies.org/"> <svg xmlns="http://www.w3.org/2000/svg" id="nasemLogo" data-name="nasemLogo" viewBox="0 0 360 66.64"> <defs/> <path d="m4.26 8.63 19.56 20.49V1.72h-3.47v18.93L.8.16v27.32h3.46V8.63zM32.55 27.48h3.83l8.15-19.2 7.93 18.99.09.21h3.82L44.62 0 32.55 27.48zM175.05 27.48h3.83l8.16-19.2 7.93 18.99.08.21h3.83L187.12 0l-12.07 27.48zM68.23 27.48h3.47V5.03h7.72V1.72H60.51v3.31h7.72v22.45zM90.08 1.72h3.46v25.76h-3.46z" class="cls-2"/> <path d="M157.14 38.67A13.39 13.39 0 1 0 143.75 52a13.21 13.21 0 0 0 13.39-13.33Zm-3.54 0a9.85 9.85 0 1 1-9.85-10 9.93 9.93 0 0 1 9.85 10Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="m146.56 8.63 19.56 20.49V1.72h-3.47v18.93L143.1.16v27.32h3.46V8.63zM210.95 1.72h-3.46v25.76h13.8v-3.31h-10.34V1.72z" class="cls-2"/> <path d="M67.41 63A13.13 13.13 0 0 0 54 76.37a13 13 0 0 0 13.26 13.24A13.48 13.48 0 0 0 75 87.1l.14-.1v-4.46l-.58.58a10.91 10.91 0 0 1-2.81 2 10 10 0 0 1-4.51 1.13c-6.74 0-9.77-5.78-9.77-9.89a9.82 9.82 0 0 1 9.74-10.11 10.44 10.44 0 0 1 4.18.94 12 12 0 0 1 3.17 2.27l.58.58v-4.41l-.13-.1a12.1 12.1 0 0 0-7.6-2.53ZM130 66.4a14 14 0 0 0-9.47-3h-5.45v25.75h5.42c4.25 0 6.88-.83 9.39-3a12.32 12.32 0 0 0 4.52-9.89A12.54 12.54 0 0 0 130 66.4Zm-2.39 17.26a10.37 10.37 0 0 1-7 2.18h-2.11V66.7h2.11c6.21 0 10.22 3.77 10.22 9.59a9.23 9.23 0 0 1-3.25 7.37Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="M119.94 65.11h15.33V61.8h-11.86v-9.07h10.38v-3.31h-10.38v-6.76h11.86v-3.32h-15.33v25.77zM201.26 65.11h15.33V61.8h-11.86v-9.07h10.37v-3.31h-10.37v-6.76h11.86v-3.32h-15.33v25.77zM158.66 58.65l-9.79-21.53-4.42 27.99h3.56l2.36-16.59 8.25 18.12 8.35-18.12 2.25 16.59h3.56l-4.14-28.03-9.98 21.57zM184.29 39.34h3.46V65.1h-3.46z" class="cls-2"/> <path d="M263.7 76.76a12.7 12.7 0 0 0-4-2.44l-2.42-1a8 8 0 0 1-2.53-1.51 2.32 2.32 0 0 1-.83-1.94c0-2.11 1.56-3.53 3.87-3.53a4.18 4.18 0 0 1 2.61.79 5.51 5.51 0 0 1 1.37 1.77l.17.34 2.8-1.85-.14-.27a7.79 7.79 0 0 0-2.56-2.91 7.16 7.16 0 0 0-4.18-1.21c-4.34 0-7.37 2.8-7.37 6.8a5.53 5.53 0 0 0 1.83 4.35 10.74 10.74 0 0 0 3.24 2l2.44 1.17a10.81 10.81 0 0 1 2.94 1.68 3.39 3.39 0 0 1 1.13 2.68A4.8 4.8 0 0 1 254 85a5.6 5.6 0 0 1-1.42-3l-.05-.39-3.4.94.07.32a9.51 9.51 0 0 0 2.43 4.61 7.73 7.73 0 0 0 5.69 2.13 7.9 7.9 0 0 0 8.28-8 6.38 6.38 0 0 0-1.9-4.85Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="M0 65.11h3.83l8.16-19.21 7.92 19 .09.21h3.83L12.07 37.63 0 65.11zM57.37 65.11h3.83l8.15-19.21 7.93 19 .08.21h3.83L69.44 37.63 57.37 65.11z" class="cls-2"/> <path d="M292 26.45a3.91 3.91 0 0 0-1.07-.14 2.94 2.94 0 0 0-1.74.52 1.71 1.71 0 0 0-.72 1.47 2.36 2.36 0 0 0 .63 1.7 9.45 9.45 0 0 0 1.9 1.38 15.52 15.52 0 0 1 1.45 1 4.28 4.28 0 0 1 .94 1.12 2.87 2.87 0 0 1 .38 1.51 3.57 3.57 0 0 1-.57 2 3.79 3.79 0 0 1-1.6 1.34 5.79 5.79 0 0 1-2.39.47 9.63 9.63 0 0 1-2.22-.25 4.15 4.15 0 0 1-1.41-.52l.39-2.49H287l.11 1.84a2.93 2.93 0 0 0 .92.39 5 5 0 0 0 1.16.14 3.39 3.39 0 0 0 2-.6 1.94 1.94 0 0 0 .83-1.68 2.25 2.25 0 0 0-.56-1.54 7.66 7.66 0 0 0-1.7-1.32c-.6-.39-1.1-.75-1.51-1.07a4.43 4.43 0 0 1-1-1.2 2.94 2.94 0 0 1-.43-1.59 3.08 3.08 0 0 1 .57-1.85 3.58 3.58 0 0 1 1.55-1.23 5.17 5.17 0 0 1 2.13-.43 7.84 7.84 0 0 1 2 .21 4 4 0 0 1 1.25.51l-.4 2.31h-.95l-.14-1.68a2.39 2.39 0 0 0-.83-.32ZM300.27 29.79a2.2 2.2 0 0 0-1.53.62 4.15 4.15 0 0 0-1.06 1.74 8 8 0 0 0-.38 2.55 4.11 4.11 0 0 0 .5 2.24 1.6 1.6 0 0 0 1.42.74 2.71 2.71 0 0 0 1.25-.27 5.56 5.56 0 0 0 1.15-.81l.35.64a4.42 4.42 0 0 1-1.31 1.08 3.78 3.78 0 0 1-2 .5 2.76 2.76 0 0 1-2.31-1 4.51 4.51 0 0 1-.83-2.9 7.35 7.35 0 0 1 .6-2.87 6 6 0 0 1 1.77-2.38 4.27 4.27 0 0 1 2.77-.95 5.7 5.7 0 0 1 1.06.11 3.31 3.31 0 0 1 .92.3l-.48 1.7-.29-.11a2.91 2.91 0 0 0-1.6-.93ZM306.13 37.31a.54.54 0 0 0 .08.33.22.22 0 0 0 .2.11.74.74 0 0 0 .39-.12 2.88 2.88 0 0 0 .5-.4l.23.42a3.34 3.34 0 0 1-.78.77 2 2 0 0 1-1.23.4 1.3 1.3 0 0 1-.89-.27.9.9 0 0 1-.26-.81c0-.3.18-1.43.46-3.41q.24-1.59.6-4.23l-1.09-.4.11-.6 2.49-.4.38.19-1.17 8.25Zm-.08-10.8a1 1 0 0 1-.27-.69 1.24 1.24 0 0 1 .34-.89 1.07 1.07 0 0 1 .82-.37.94.94 0 0 1 .7.27.93.93 0 0 1 .26.66 1.35 1.35 0 0 1-.33.92 1.14 1.14 0 0 1-.88.38.9.9 0 0 1-.64-.28ZM315.15 38.24a4.26 4.26 0 0 1-2.23.6 3.45 3.45 0 0 1-2-.53 3.23 3.23 0 0 1-1.15-1.41 4.83 4.83 0 0 1-.36-1.87 7.38 7.38 0 0 1 .67-3.14 5.91 5.91 0 0 1 1.92-2.33 4.33 4.33 0 0 1 2.62-.88 2.64 2.64 0 0 1 1.88.6 2.06 2.06 0 0 1 .63 1.56 2.66 2.66 0 0 1-1 2.11 6.62 6.62 0 0 1-2.37 1.28 9.48 9.48 0 0 1-2.46.45 3.85 3.85 0 0 0 .42 2.13 1.68 1.68 0 0 0 1.6.91 3.61 3.61 0 0 0 1.54-.36 4.41 4.41 0 0 0 1.39-1l.38.65a4.39 4.39 0 0 1-1.48 1.23Zm-2.41-8a4.82 4.82 0 0 0-1 1.65 7.41 7.41 0 0 0-.44 1.94 5.85 5.85 0 0 0 1.86-.39 4.34 4.34 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.33-1.05 1.14 1.14 0 0 0-.88-.35 2 2 0 0 0-1.46.67ZM323.42 29.32a3.62 3.62 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.93 2.93 0 0 1 .53 1.95 16.77 16.77 0 0 1-.39 2.68l-.07.4-.14.85a12.6 12.6 0 0 0-.28 2.12c0 .32.08.47.23.47a1.35 1.35 0 0 0 .82-.44l.21.46a1.7 1.7 0 0 1-.36.4 2.81 2.81 0 0 1-.7.45 1.92 1.92 0 0 1-.9.2c-.74 0-1.11-.35-1.11-1.07a14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.63.21-1.18.28-1.66a9 9 0 0 0 .1-1.32 2 2 0 0 0-.23-1.12 1 1 0 0 0-.84-.3 2.63 2.63 0 0 0-1.57.69 6.6 6.6 0 0 0-1.48 1.67l-.86 6.34H319l1.13-8.49-.86-.42.14-.67 2.14-.39.37.2-.23 2.1a6.56 6.56 0 0 1 1.73-1.66ZM334.3 29.79a2.22 2.22 0 0 0-1.53.62 4.14 4.14 0 0 0-1 1.74 7.68 7.68 0 0 0-.38 2.55 4.11 4.11 0 0 0 .49 2.24 1.61 1.61 0 0 0 1.42.74 2.78 2.78 0 0 0 1.26-.27 5.52 5.52 0 0 0 1.14-.81l.36.64a4.56 4.56 0 0 1-1.31 1.08 3.85 3.85 0 0 1-2 .5 2.79 2.79 0 0 1-2.31-1 4.57 4.57 0 0 1-.83-2.9 7.19 7.19 0 0 1 .61-2.87 5.94 5.94 0 0 1 1.77-2.38 4.24 4.24 0 0 1 2.76-.95 5.7 5.7 0 0 1 1.06.11 3.2 3.2 0 0 1 .92.3l-.47 1.7-.3-.11a2.91 2.91 0 0 0-1.66-.93ZM343.32 38.24a4.26 4.26 0 0 1-2.23.6 3.42 3.42 0 0 1-1.95-.53A3.17 3.17 0 0 1 338 36.9a4.66 4.66 0 0 1-.37-1.87 7.53 7.53 0 0 1 .67-3.14 6 6 0 0 1 1.86-2.33 4.36 4.36 0 0 1 2.63-.88 2.65 2.65 0 0 1 1.88.6 2.06 2.06 0 0 1 .63 1.56 2.69 2.69 0 0 1-1 2.11 6.66 6.66 0 0 1-2.38 1.28 9.33 9.33 0 0 1-2.46.45 4 4 0 0 0 .42 2.13 1.68 1.68 0 0 0 1.6.91 3.65 3.65 0 0 0 1.55-.36 4.37 4.37 0 0 0 1.38-1l.39.65a4.43 4.43 0 0 1-1.48 1.23Zm-2.4-8a4.67 4.67 0 0 0-1 1.65 7 7 0 0 0-.44 1.94 5.79 5.79 0 0 0 1.85-.39 4.25 4.25 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.32-1.05 1.17 1.17 0 0 0-.89-.35 1.93 1.93 0 0 0-1.45.67ZM351.91 29.91a3.08 3.08 0 0 0-1.24-.24 2.15 2.15 0 0 0-1.3.35 1.18 1.18 0 0 0-.49 1 1.54 1.54 0 0 0 .4 1.06 6.17 6.17 0 0 0 1.13 1c.42.33.76.61 1 .85a3.35 3.35 0 0 1 .67.87 2.38 2.38 0 0 1 .3 1.2 2.4 2.4 0 0 1-1 2.07 4.21 4.21 0 0 1-2.53.73 7.55 7.55 0 0 1-1.57-.17 2.73 2.73 0 0 1-1-.39l.29-1.21a3.77 3.77 0 0 0 1 .5 3.94 3.94 0 0 0 1.36.24 2.47 2.47 0 0 0 1.31-.33 1.14 1.14 0 0 0 .55-1.06 1.58 1.58 0 0 0-.43-1.09 8 8 0 0 0-1.26-1 3.53 3.53 0 0 1-1.81-2.86 2.43 2.43 0 0 1 .46-1.45 2.92 2.92 0 0 1 1.25-1 4.48 4.48 0 0 1 1.75-.34 7.2 7.2 0 0 1 1.29.12 4.56 4.56 0 0 1 .92.22l-.3 1.33a1.83 1.83 0 0 0-.75-.4ZM287 50.81h8.32l-.12 2.79h-1l-.28-1.84-4-.14-.67 5 4.07-.14-.12 1.32-4-.14-.81 5.34 4.95-.21 1-2.12 1 .19-.71 3h-9.33l.07-.68 1.32-.21 1.4-11.19-1.19-.2ZM301.65 54.52a3.52 3.52 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.89 2.89 0 0 1 .53 1.94 16.85 16.85 0 0 1-.39 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.07.47.23.47a1.41 1.41 0 0 0 .82-.44l.14.49a1.76 1.76 0 0 1-.37.4 3.17 3.17 0 0 1-.69.45 2 2 0 0 1-.9.2A1 1 0 0 1 303 63a14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.64.21-1.19.28-1.66a9.18 9.18 0 0 0 .1-1.33 2 2 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33h-1.67l1.13-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.56 6.56 0 0 1 1.68-1.71ZM313.74 66.64a4.23 4.23 0 0 1-1.77 1.2 5.86 5.86 0 0 1-2 .37 7.37 7.37 0 0 1-1.13-.09 4.89 4.89 0 0 1-.87-.22l.37-1.42a6.44 6.44 0 0 0 1 .39 4.29 4.29 0 0 0 1.22.18 2.62 2.62 0 0 0 1.66-.43 2.58 2.58 0 0 0 .71-1.62l.33-2.33a4.25 4.25 0 0 1-1.18 1 3 3 0 0 1-1.52.39 2.52 2.52 0 0 1-2.17-1 4.5 4.5 0 0 1-.72-2.68 7.33 7.33 0 0 1 .66-3 6.45 6.45 0 0 1 1.95-2.49 4.73 4.73 0 0 1 3-1 5.1 5.1 0 0 1 1 .11 5.19 5.19 0 0 1 .91.28l1.09-.39-1.58 10.63a3.73 3.73 0 0 1-.96 2.12ZM313 54.81a2.66 2.66 0 0 0-2 .85 5.43 5.43 0 0 0-1.22 2.11 7.91 7.91 0 0 0-.4 2.46 3.81 3.81 0 0 0 .42 2 1.31 1.31 0 0 0 1.17.71 2.42 2.42 0 0 0 1.37-.44 3.83 3.83 0 0 0 1-1l.93-6.28a2.13 2.13 0 0 0-1.27-.41ZM320.16 62.51a.56.56 0 0 0 .08.32.24.24 0 0 0 .2.12.8.8 0 0 0 .4-.12 4.16 4.16 0 0 0 .5-.41l.23.42a3.57 3.57 0 0 1-.78.78 2 2 0 0 1-1.24.4 1.32 1.32 0 0 1-.89-.27.92.92 0 0 1-.25-.82c0-.29.18-1.43.46-3.4.16-1.07.36-2.48.59-4.23l-1.09-.4.11-.6 2.49-.4.39.19-1.18 8.25Zm-.08-10.81a.93.93 0 0 1-.27-.68 1.25 1.25 0 0 1 .35-.89 1 1 0 0 1 .81-.37.93.93 0 0 1 .7.27.9.9 0 0 1 .27.66 1.36 1.36 0 0 1-.34.92 1.11 1.11 0 0 1-.88.37.87.87 0 0 1-.64-.28ZM328.56 54.52a3.54 3.54 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.89 2.89 0 0 1 .53 1.94 16.85 16.85 0 0 1-.39 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.08.47.23.47a1.41 1.41 0 0 0 .82-.44l.23.49a1.76 1.76 0 0 1-.37.4 3.17 3.17 0 0 1-.69.45 2 2 0 0 1-.9.2 1 1 0 0 1-1.16-1.05 14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.64.21-1.19.28-1.66a9.18 9.18 0 0 0 .1-1.33 2 2 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33h-1.67l1.13-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.56 6.56 0 0 1 1.71-1.71ZM340.41 63.44a4.26 4.26 0 0 1-2.23.6 3.35 3.35 0 0 1-1.95-.54 3 3 0 0 1-1.15-1.4 4.66 4.66 0 0 1-.37-1.87 7.38 7.38 0 0 1 .67-3.14 5.91 5.91 0 0 1 1.86-2.33 4.29 4.29 0 0 1 2.62-.88 2.62 2.62 0 0 1 1.88.6 2 2 0 0 1 .63 1.56 2.63 2.63 0 0 1-1 2.1 6.47 6.47 0 0 1-2.37 1.28 9.38 9.38 0 0 1-2.46.46A4 4 0 0 0 337 62a1.69 1.69 0 0 0 1.6.9 3.61 3.61 0 0 0 1.54-.36 4.27 4.27 0 0 0 1.39-1l.39.65a4.68 4.68 0 0 1-1.51 1.25Zm-2.4-8a4.53 4.53 0 0 0-1 1.65 6.69 6.69 0 0 0-.44 1.94 6.13 6.13 0 0 0 1.85-.39 4.25 4.25 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.55 1.55 0 0 0-.32-1.05 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.46.68ZM349.39 63.44a4.26 4.26 0 0 1-2.23.6 3.35 3.35 0 0 1-2-.54 3.09 3.09 0 0 1-1.15-1.4 4.66 4.66 0 0 1-.37-1.87 7.38 7.38 0 0 1 .68-3.14 5.91 5.91 0 0 1 1.86-2.33 4.25 4.25 0 0 1 2.62-.88 2.65 2.65 0 0 1 1.88.6 2 2 0 0 1 .63 1.56 2.66 2.66 0 0 1-1 2.1 6.51 6.51 0 0 1-2.31 1.28 9.38 9.38 0 0 1-2.46.46 3.94 3.94 0 0 0 .43 2.13 1.66 1.66 0 0 0 1.59.9 3.65 3.65 0 0 0 1.55-.36 4.34 4.34 0 0 0 1.38-1l.39.65a4.68 4.68 0 0 1-1.49 1.24Zm-2.4-8a4.67 4.67 0 0 0-1 1.65 7 7 0 0 0-.44 1.94 6.07 6.07 0 0 0 1.85-.39 4.38 4.38 0 0 0 1.61-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.33-1.05 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.44.68ZM356.34 55.24a4.21 4.21 0 0 1 1.13-1 2.81 2.81 0 0 1 1.4-.38.93.93 0 0 1 .56.12l-.37 1.86a1.64 1.64 0 0 0-.86-.18 2.68 2.68 0 0 0-1.48.5 3.31 3.31 0 0 0-1.14 1.4l-.8 6.24h-1.69l1.11-8.45-.88-.44.12-.65 2.07-.4.35.21-.12 1.56-.08.7a3.9 3.9 0 0 1 .68-1.09ZM362.32 62.51a.56.56 0 0 0 .08.32.23.23 0 0 0 .2.12.8.8 0 0 0 .4-.12 4.16 4.16 0 0 0 .5-.41l.22.42a3.18 3.18 0 0 1-.78.78 2 2 0 0 1-1.23.4 1.3 1.3 0 0 1-.89-.27.92.92 0 0 1-.25-.82c0-.29.17-1.43.45-3.4.17-1.07.36-2.48.6-4.23l-1.09-.4.11-.6 2.49-.4.38.19-1.17 8.25Zm-.08-10.81a.93.93 0 0 1-.24-.7 1.24 1.24 0 0 1 .34-.89 1.07 1.07 0 0 1 .82-.37.94.94 0 0 1 .7.27.93.93 0 0 1 .26.66 1.35 1.35 0 0 1-.33.92 1.12 1.12 0 0 1-.88.37.86.86 0 0 1-.67-.26ZM370.36 54.52a3.55 3.55 0 0 1 2-.64 1.91 1.91 0 0 1 1.51.6 2.89 2.89 0 0 1 .52 1.94 17.7 17.7 0 0 1-.38 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.07.47.22.47a1.4 1.4 0 0 0 .83-.44l.21.46a1.76 1.76 0 0 1-.37.4 3.74 3.74 0 0 1-.69.45 2.13 2.13 0 0 1-.91.2 1 1 0 0 1-1.1-1.07 14.1 14.1 0 0 1 .31-2.14l.16-.84.07-.42c.12-.64.21-1.19.28-1.66a9.27 9.27 0 0 0 .11-1.33 1.88 1.88 0 0 0-.24-1.11 1 1 0 0 0-.83-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33H366l1.12-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.72 6.72 0 0 1 1.68-1.66ZM382.46 66.64a4.23 4.23 0 0 1-1.77 1.2 5.87 5.87 0 0 1-2 .37 7.34 7.34 0 0 1-1.12-.09 4.82 4.82 0 0 1-.88-.22l.37-1.42a6.58 6.58 0 0 0 1 .39 4.24 4.24 0 0 0 1.22.18 2.57 2.57 0 0 0 1.65-.43 2.53 2.53 0 0 0 .72-1.62l.33-2.33a4.25 4.25 0 0 1-1.18 1 3.07 3.07 0 0 1-1.52.39 2.52 2.52 0 0 1-2.17-1 4.5 4.5 0 0 1-.73-2.68 7.35 7.35 0 0 1 .67-3 6.27 6.27 0 0 1 1.95-2.51 4.71 4.71 0 0 1 3-1 5.1 5.1 0 0 1 1 .11 5.47 5.47 0 0 1 .91.28l1.09-.39-1.58 10.63a3.86 3.86 0 0 1-.96 2.14Zm-.7-11.83a2.62 2.62 0 0 0-2 .85 5.31 5.31 0 0 0-1.22 2.11 7.93 7.93 0 0 0-.41 2.46 3.7 3.7 0 0 0 .43 2 1.31 1.31 0 0 0 1.17.71 2.42 2.42 0 0 0 1.37-.44 3.79 3.79 0 0 0 1-1l.93-6.28a2.08 2.08 0 0 0-1.27-.41ZM295.85 80.3q1.44-3.12 1.89-4.29h3.76l-.07.75-1.56.19-.34 11.2 1.49.21-.02.64h-4.82l.07-.68 1.6-.21.4-7.76.23-3.56L297.1 80l-4.1 9.08-.94.12-1.44-9-.55-3.33-.71 3.4-1.88 7.9 1.77.21-.07.68h-4.53l.07-.68 1.46-.21L289 77l-1.62-.19.11-.75h4.07l1.35 8.56.23 1.79.77-1.76q.49-1.22 1.94-4.35ZM308 88.64a4.35 4.35 0 0 1-2.23.59 3.38 3.38 0 0 1-2-.53 3.26 3.26 0 0 1-1.15-1.4 4.87 4.87 0 0 1-.36-1.87 7.34 7.34 0 0 1 .67-3.14 6 6 0 0 1 1.91-2.29 4.32 4.32 0 0 1 2.62-.87 2.63 2.63 0 0 1 1.88.59 2.06 2.06 0 0 1 .63 1.56 2.65 2.65 0 0 1-1 2.11 6.62 6.62 0 0 1-2.37 1.28 10 10 0 0 1-2.46.46 3.89 3.89 0 0 0 .42 2.13 1.69 1.69 0 0 0 1.6.9 3.61 3.61 0 0 0 1.54-.36 4.3 4.3 0 0 0 1.39-1l.38.65a4.52 4.52 0 0 1-1.47 1.19Zm-2.41-8a4.68 4.68 0 0 0-1 1.64 7.5 7.5 0 0 0-.44 1.94 6.11 6.11 0 0 0 1.86-.38 4.61 4.61 0 0 0 1.6-1 2 2 0 0 0 .65-1.47 1.54 1.54 0 0 0-.32-1 1.15 1.15 0 0 0-.89-.36 2 2 0 0 0-1.45.64ZM318.78 87.71c0 .29.09.44.26.44a1.42 1.42 0 0 0 .81-.44l.24.44a3.31 3.31 0 0 1-.76.71 1.9 1.9 0 0 1-1.15.36c-.69 0-1-.38-1-1.13v-.38a4.57 4.57 0 0 1-1.23 1.07 3.18 3.18 0 0 1-1.65.44 2.46 2.46 0 0 1-2.14-1 4.54 4.54 0 0 1-.72-2.68 7.24 7.24 0 0 1 .67-3 6.3 6.3 0 0 1 1.89-2.48 4.67 4.67 0 0 1 3-1 4.42 4.42 0 0 1 1.31.21l.48-3.16-1.16-.3.14-.61 2.47-.32.39.23-1.83 12.4a.59.59 0 0 0-.02.2Zm-2-7.7a2.62 2.62 0 0 0-2 .85 5.14 5.14 0 0 0-1.21 2.1 8 8 0 0 0-.4 2.47 3.83 3.83 0 0 0 .42 2 1.27 1.27 0 0 0 1.16.7 2.39 2.39 0 0 0 1.4-.46 3.7 3.7 0 0 0 1-1l.95-6.19a2.2 2.2 0 0 0-1.32-.48ZM324.44 87.71a.56.56 0 0 0 .08.32.25.25 0 0 0 .21.12.79.79 0 0 0 .39-.13 3.18 3.18 0 0 0 .5-.4l.23.42a3.35 3.35 0 0 1-.78.77 2 2 0 0 1-1.24.41 1.34 1.34 0 0 1-.89-.28.92.92 0 0 1-.25-.81c0-.29.18-1.43.46-3.41.16-1.06.36-2.47.59-4.22l-1.08-.41.1-.59 2.49-.41.39.2-1.18 8.24Zm-.07-10.81a1 1 0 0 1-.28-.68 1.29 1.29 0 0 1 .35-.9 1.1 1.1 0 0 1 .81-.37.92.92 0 0 1 1 .93 1.29 1.29 0 0 1-.34.92 1.11 1.11 0 0 1-.87.38.84.84 0 0 1-.67-.28ZM332.76 80.18a2.22 2.22 0 0 0-1.53.62 4.15 4.15 0 0 0-1.06 1.74 8 8 0 0 0-.37 2.55 4.09 4.09 0 0 0 .49 2.24 1.59 1.59 0 0 0 1.42.75 2.71 2.71 0 0 0 1.29-.28 5.56 5.56 0 0 0 1.15-.81l.35.65a4.84 4.84 0 0 1-1.3 1.08 3.94 3.94 0 0 1-2 .5 2.78 2.78 0 0 1-2.31-1 4.49 4.49 0 0 1-.83-2.9 7.35 7.35 0 0 1 .6-2.87 6.1 6.1 0 0 1 1.72-2.45 4.26 4.26 0 0 1 2.77-.94 5.66 5.66 0 0 1 1.06.1 3.31 3.31 0 0 1 .92.3l-.47 1.7-.3-.1a2.88 2.88 0 0 0-1.6-.88ZM338.88 87.71a.56.56 0 0 0 .12.29.23.23 0 0 0 .2.12.82.82 0 0 0 .4-.13 3.18 3.18 0 0 0 .5-.4l.23.42a3.56 3.56 0 0 1-.78.77 2 2 0 0 1-1.24.41 1.32 1.32 0 0 1-.89-.28.9.9 0 0 1-.25-.81c0-.29.18-1.43.46-3.41.16-1.06.36-2.47.59-4.22l-1.09-.41.11-.59 2.49-.41.39.2-1.18 8.24Zm-.08-10.81a1 1 0 0 1-.27-.68 1.29 1.29 0 0 1 .35-.9 1.07 1.07 0 0 1 .81-.37.92.92 0 0 1 1 .93 1.34 1.34 0 0 1-.34.92 1.11 1.11 0 0 1-.88.38.87.87 0 0 1-.67-.28ZM347.19 79.72a3.54 3.54 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.59 2.93 2.93 0 0 1 .53 2 16.77 16.77 0 0 1-.39 2.68l-.07.41-.14.84a12.6 12.6 0 0 0-.28 2.12c0 .32.08.48.23.48a1.38 1.38 0 0 0 .82-.44l.21.45a1.91 1.91 0 0 1-.36.41 3.11 3.11 0 0 1-.7.44 1.93 1.93 0 0 1-.9.21 1 1 0 0 1-1.11-1.07 14.24 14.24 0 0 1 .32-2.14l.16-.85.07-.42c.11-.63.21-1.18.28-1.66a9 9 0 0 0 .1-1.32 1.87 1.87 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.66 2.66 0 0 0-1.57.69 6.48 6.48 0 0 0-1.48 1.68l-.89 6.24h-1.66l1.12-8.49-.86-.42.14-.67 2.14-.38.37.19-.23 2.1a6.52 6.52 0 0 1 1.71-1.61ZM359 88.64a4.35 4.35 0 0 1-2.23.59 3.34 3.34 0 0 1-1.95-.53 3.19 3.19 0 0 1-1.15-1.4 4.69 4.69 0 0 1-.37-1.87 7.34 7.34 0 0 1 .67-3.14 6 6 0 0 1 1.9-2.29 4.35 4.35 0 0 1 2.63-.87 2.61 2.61 0 0 1 1.87.59 2 2 0 0 1 .64 1.56 2.68 2.68 0 0 1-1 2.11 6.75 6.75 0 0 1-2.38 1.28 9.81 9.81 0 0 1-2.46.46 4.08 4.08 0 0 0 .42 2.13 1.69 1.69 0 0 0 1.6.9 3.65 3.65 0 0 0 1.55-.36 4.37 4.37 0 0 0 1.38-1l.39.65a4.57 4.57 0 0 1-1.51 1.19Zm-2.4-8a4.54 4.54 0 0 0-1 1.64 6.76 6.76 0 0 0-.44 1.94 6 6 0 0 0 1.85-.38 4.52 4.52 0 0 0 1.6-1 2 2 0 0 0 .65-1.47 1.49 1.49 0 0 0-.32-1 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.41.64Z" class="cls-2" transform="translate(-24.97 -24.05)"/> </svg> </a></div> <div class="global-nav__controls"><button class="controls__search" style="display: none;"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path d="M13.8 21.6a7.8 7.8 0 110-15.6 7.8 7.8 0 010 15.6zm0-2a5.8 5.8 0 100-11.6 5.8 5.8 0 000 11.6z"></path><path d="M17.899 19.325l1.414-1.414 4.322 4.322-1.414 1.414z"></path></svg></button> <button class="controls__close" onclick="toggleSearch()"><span class="menus__close-label"> Close </span> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M20.728 8l1.818 1.818-5.455 5.455 5.455 5.455-1.818 1.818-5.455-5.455-5.455 5.455L8 20.728l5.455-5.455L8 9.818 9.818 8l5.455 5.455z"></path></svg></button></div></div> <div class="mega-menu__content"> <div class="search-form sf-top" > <div class="search-form__container"> <label for="inputsearch"> Search for Publications Only </label> <form id="searchFormNAP" action="/search" enctype="application/x-www-form-urlencoded"> <input id="inputsearch" name="term" type="text" placeholder="Type a title, topic, ISBN, or DOI and hit enter"> <input type="hidden" name="v:project" value="uweb_proj_ext"> <button class="search-form__submit"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""> <path d="M13.8 21.6a7.8 7.8 0 110-15.6 7.8 7.8 0 010 15.6zm0-2a5.8 5.8 0 100-11.6 5.8 5.8 0 000 11.6z"></path> <path d="M17.899 19.325l1.414-1.414 4.322 4.322-1.414 1.414z"></path> </svg> </button> </form> </div> <!----></div> <div class="search-form"> <div class="search-form__container"> <label for="globalSearch"> Search All</label> <form id="searchForm" action="https://sparck.nationalacademies.org/vivisimo/cgi-bin/query-meta" enctype="application/x-www-form-urlencoded"> <input id="globalSearch" name="query" type="text" placeholder="Type anything and hit enter"> <input type="hidden" name="v:project" value="uweb_proj_ext"> <button class="search-form__submit"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""> <path d="M13.8 21.6a7.8 7.8 0 110-15.6 7.8 7.8 0 010 15.6zm0-2a5.8 5.8 0 100-11.6 5.8 5.8 0 000 11.6z"></path> <path d="M17.899 19.325l1.414-1.414 4.322 4.322-1.414 1.414z"></path> </svg> </button> </form> </div> <!----></div> <!----></div> </div> <div tabindex="0" aria-hidden="true"></div> </div> <!--Mega menu end --> <!--Mobile menu start --> <div aria-hidden="true" class="mega-menu" id="mobile-menu" style="display: none;"> <div data-lock=""> <div class="mega-menu__header"> <div class="global-nav__logo" style="padding-left: 10px;"> <a class="navbar-brand js-scroll-trigger" href="https://www.nationalacademies.org/"> <svg xmlns="http://www.w3.org/2000/svg" id="nasemLogo" data-name="nasemLogo" viewBox="0 0 360 66.64"> <defs/> <path d="m4.26 8.63 19.56 20.49V1.72h-3.47v18.93L.8.16v27.32h3.46V8.63zM32.55 27.48h3.83l8.15-19.2 7.93 18.99.09.21h3.82L44.62 0 32.55 27.48zM175.05 27.48h3.83l8.16-19.2 7.93 18.99.08.21h3.83L187.12 0l-12.07 27.48zM68.23 27.48h3.47V5.03h7.72V1.72H60.51v3.31h7.72v22.45zM90.08 1.72h3.46v25.76h-3.46z" class="cls-2"/> <path d="M157.14 38.67A13.39 13.39 0 1 0 143.75 52a13.21 13.21 0 0 0 13.39-13.33Zm-3.54 0a9.85 9.85 0 1 1-9.85-10 9.93 9.93 0 0 1 9.85 10Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="m146.56 8.63 19.56 20.49V1.72h-3.47v18.93L143.1.16v27.32h3.46V8.63zM210.95 1.72h-3.46v25.76h13.8v-3.31h-10.34V1.72z" class="cls-2"/> <path d="M67.41 63A13.13 13.13 0 0 0 54 76.37a13 13 0 0 0 13.26 13.24A13.48 13.48 0 0 0 75 87.1l.14-.1v-4.46l-.58.58a10.91 10.91 0 0 1-2.81 2 10 10 0 0 1-4.51 1.13c-6.74 0-9.77-5.78-9.77-9.89a9.82 9.82 0 0 1 9.74-10.11 10.44 10.44 0 0 1 4.18.94 12 12 0 0 1 3.17 2.27l.58.58v-4.41l-.13-.1a12.1 12.1 0 0 0-7.6-2.53ZM130 66.4a14 14 0 0 0-9.47-3h-5.45v25.75h5.42c4.25 0 6.88-.83 9.39-3a12.32 12.32 0 0 0 4.52-9.89A12.54 12.54 0 0 0 130 66.4Zm-2.39 17.26a10.37 10.37 0 0 1-7 2.18h-2.11V66.7h2.11c6.21 0 10.22 3.77 10.22 9.59a9.23 9.23 0 0 1-3.25 7.37Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="M119.94 65.11h15.33V61.8h-11.86v-9.07h10.38v-3.31h-10.38v-6.76h11.86v-3.32h-15.33v25.77zM201.26 65.11h15.33V61.8h-11.86v-9.07h10.37v-3.31h-10.37v-6.76h11.86v-3.32h-15.33v25.77zM158.66 58.65l-9.79-21.53-4.42 27.99h3.56l2.36-16.59 8.25 18.12 8.35-18.12 2.25 16.59h3.56l-4.14-28.03-9.98 21.57zM184.29 39.34h3.46V65.1h-3.46z" class="cls-2"/> <path d="M263.7 76.76a12.7 12.7 0 0 0-4-2.44l-2.42-1a8 8 0 0 1-2.53-1.51 2.32 2.32 0 0 1-.83-1.94c0-2.11 1.56-3.53 3.87-3.53a4.18 4.18 0 0 1 2.61.79 5.51 5.51 0 0 1 1.37 1.77l.17.34 2.8-1.85-.14-.27a7.79 7.79 0 0 0-2.56-2.91 7.16 7.16 0 0 0-4.18-1.21c-4.34 0-7.37 2.8-7.37 6.8a5.53 5.53 0 0 0 1.83 4.35 10.74 10.74 0 0 0 3.24 2l2.44 1.17a10.81 10.81 0 0 1 2.94 1.68 3.39 3.39 0 0 1 1.13 2.68A4.8 4.8 0 0 1 254 85a5.6 5.6 0 0 1-1.42-3l-.05-.39-3.4.94.07.32a9.51 9.51 0 0 0 2.43 4.61 7.73 7.73 0 0 0 5.69 2.13 7.9 7.9 0 0 0 8.28-8 6.38 6.38 0 0 0-1.9-4.85Z" class="cls-2" transform="translate(-24.97 -24.05)"/> <path d="M0 65.11h3.83l8.16-19.21 7.92 19 .09.21h3.83L12.07 37.63 0 65.11zM57.37 65.11h3.83l8.15-19.21 7.93 19 .08.21h3.83L69.44 37.63 57.37 65.11z" class="cls-2"/> <path d="M292 26.45a3.91 3.91 0 0 0-1.07-.14 2.94 2.94 0 0 0-1.74.52 1.71 1.71 0 0 0-.72 1.47 2.36 2.36 0 0 0 .63 1.7 9.45 9.45 0 0 0 1.9 1.38 15.52 15.52 0 0 1 1.45 1 4.28 4.28 0 0 1 .94 1.12 2.87 2.87 0 0 1 .38 1.51 3.57 3.57 0 0 1-.57 2 3.79 3.79 0 0 1-1.6 1.34 5.79 5.79 0 0 1-2.39.47 9.63 9.63 0 0 1-2.22-.25 4.15 4.15 0 0 1-1.41-.52l.39-2.49H287l.11 1.84a2.93 2.93 0 0 0 .92.39 5 5 0 0 0 1.16.14 3.39 3.39 0 0 0 2-.6 1.94 1.94 0 0 0 .83-1.68 2.25 2.25 0 0 0-.56-1.54 7.66 7.66 0 0 0-1.7-1.32c-.6-.39-1.1-.75-1.51-1.07a4.43 4.43 0 0 1-1-1.2 2.94 2.94 0 0 1-.43-1.59 3.08 3.08 0 0 1 .57-1.85 3.58 3.58 0 0 1 1.55-1.23 5.17 5.17 0 0 1 2.13-.43 7.84 7.84 0 0 1 2 .21 4 4 0 0 1 1.25.51l-.4 2.31h-.95l-.14-1.68a2.39 2.39 0 0 0-.83-.32ZM300.27 29.79a2.2 2.2 0 0 0-1.53.62 4.15 4.15 0 0 0-1.06 1.74 8 8 0 0 0-.38 2.55 4.11 4.11 0 0 0 .5 2.24 1.6 1.6 0 0 0 1.42.74 2.71 2.71 0 0 0 1.25-.27 5.56 5.56 0 0 0 1.15-.81l.35.64a4.42 4.42 0 0 1-1.31 1.08 3.78 3.78 0 0 1-2 .5 2.76 2.76 0 0 1-2.31-1 4.51 4.51 0 0 1-.83-2.9 7.35 7.35 0 0 1 .6-2.87 6 6 0 0 1 1.77-2.38 4.27 4.27 0 0 1 2.77-.95 5.7 5.7 0 0 1 1.06.11 3.31 3.31 0 0 1 .92.3l-.48 1.7-.29-.11a2.91 2.91 0 0 0-1.6-.93ZM306.13 37.31a.54.54 0 0 0 .08.33.22.22 0 0 0 .2.11.74.74 0 0 0 .39-.12 2.88 2.88 0 0 0 .5-.4l.23.42a3.34 3.34 0 0 1-.78.77 2 2 0 0 1-1.23.4 1.3 1.3 0 0 1-.89-.27.9.9 0 0 1-.26-.81c0-.3.18-1.43.46-3.41q.24-1.59.6-4.23l-1.09-.4.11-.6 2.49-.4.38.19-1.17 8.25Zm-.08-10.8a1 1 0 0 1-.27-.69 1.24 1.24 0 0 1 .34-.89 1.07 1.07 0 0 1 .82-.37.94.94 0 0 1 .7.27.93.93 0 0 1 .26.66 1.35 1.35 0 0 1-.33.92 1.14 1.14 0 0 1-.88.38.9.9 0 0 1-.64-.28ZM315.15 38.24a4.26 4.26 0 0 1-2.23.6 3.45 3.45 0 0 1-2-.53 3.23 3.23 0 0 1-1.15-1.41 4.83 4.83 0 0 1-.36-1.87 7.38 7.38 0 0 1 .67-3.14 5.91 5.91 0 0 1 1.92-2.33 4.33 4.33 0 0 1 2.62-.88 2.64 2.64 0 0 1 1.88.6 2.06 2.06 0 0 1 .63 1.56 2.66 2.66 0 0 1-1 2.11 6.62 6.62 0 0 1-2.37 1.28 9.48 9.48 0 0 1-2.46.45 3.85 3.85 0 0 0 .42 2.13 1.68 1.68 0 0 0 1.6.91 3.61 3.61 0 0 0 1.54-.36 4.41 4.41 0 0 0 1.39-1l.38.65a4.39 4.39 0 0 1-1.48 1.23Zm-2.41-8a4.82 4.82 0 0 0-1 1.65 7.41 7.41 0 0 0-.44 1.94 5.85 5.85 0 0 0 1.86-.39 4.34 4.34 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.33-1.05 1.14 1.14 0 0 0-.88-.35 2 2 0 0 0-1.46.67ZM323.42 29.32a3.62 3.62 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.93 2.93 0 0 1 .53 1.95 16.77 16.77 0 0 1-.39 2.68l-.07.4-.14.85a12.6 12.6 0 0 0-.28 2.12c0 .32.08.47.23.47a1.35 1.35 0 0 0 .82-.44l.21.46a1.7 1.7 0 0 1-.36.4 2.81 2.81 0 0 1-.7.45 1.92 1.92 0 0 1-.9.2c-.74 0-1.11-.35-1.11-1.07a14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.63.21-1.18.28-1.66a9 9 0 0 0 .1-1.32 2 2 0 0 0-.23-1.12 1 1 0 0 0-.84-.3 2.63 2.63 0 0 0-1.57.69 6.6 6.6 0 0 0-1.48 1.67l-.86 6.34H319l1.13-8.49-.86-.42.14-.67 2.14-.39.37.2-.23 2.1a6.56 6.56 0 0 1 1.73-1.66ZM334.3 29.79a2.22 2.22 0 0 0-1.53.62 4.14 4.14 0 0 0-1 1.74 7.68 7.68 0 0 0-.38 2.55 4.11 4.11 0 0 0 .49 2.24 1.61 1.61 0 0 0 1.42.74 2.78 2.78 0 0 0 1.26-.27 5.52 5.52 0 0 0 1.14-.81l.36.64a4.56 4.56 0 0 1-1.31 1.08 3.85 3.85 0 0 1-2 .5 2.79 2.79 0 0 1-2.31-1 4.57 4.57 0 0 1-.83-2.9 7.19 7.19 0 0 1 .61-2.87 5.94 5.94 0 0 1 1.77-2.38 4.24 4.24 0 0 1 2.76-.95 5.7 5.7 0 0 1 1.06.11 3.2 3.2 0 0 1 .92.3l-.47 1.7-.3-.11a2.91 2.91 0 0 0-1.66-.93ZM343.32 38.24a4.26 4.26 0 0 1-2.23.6 3.42 3.42 0 0 1-1.95-.53A3.17 3.17 0 0 1 338 36.9a4.66 4.66 0 0 1-.37-1.87 7.53 7.53 0 0 1 .67-3.14 6 6 0 0 1 1.86-2.33 4.36 4.36 0 0 1 2.63-.88 2.65 2.65 0 0 1 1.88.6 2.06 2.06 0 0 1 .63 1.56 2.69 2.69 0 0 1-1 2.11 6.66 6.66 0 0 1-2.38 1.28 9.33 9.33 0 0 1-2.46.45 4 4 0 0 0 .42 2.13 1.68 1.68 0 0 0 1.6.91 3.65 3.65 0 0 0 1.55-.36 4.37 4.37 0 0 0 1.38-1l.39.65a4.43 4.43 0 0 1-1.48 1.23Zm-2.4-8a4.67 4.67 0 0 0-1 1.65 7 7 0 0 0-.44 1.94 5.79 5.79 0 0 0 1.85-.39 4.25 4.25 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.32-1.05 1.17 1.17 0 0 0-.89-.35 1.93 1.93 0 0 0-1.45.67ZM351.91 29.91a3.08 3.08 0 0 0-1.24-.24 2.15 2.15 0 0 0-1.3.35 1.18 1.18 0 0 0-.49 1 1.54 1.54 0 0 0 .4 1.06 6.17 6.17 0 0 0 1.13 1c.42.33.76.61 1 .85a3.35 3.35 0 0 1 .67.87 2.38 2.38 0 0 1 .3 1.2 2.4 2.4 0 0 1-1 2.07 4.21 4.21 0 0 1-2.53.73 7.55 7.55 0 0 1-1.57-.17 2.73 2.73 0 0 1-1-.39l.29-1.21a3.77 3.77 0 0 0 1 .5 3.94 3.94 0 0 0 1.36.24 2.47 2.47 0 0 0 1.31-.33 1.14 1.14 0 0 0 .55-1.06 1.58 1.58 0 0 0-.43-1.09 8 8 0 0 0-1.26-1 3.53 3.53 0 0 1-1.81-2.86 2.43 2.43 0 0 1 .46-1.45 2.92 2.92 0 0 1 1.25-1 4.48 4.48 0 0 1 1.75-.34 7.2 7.2 0 0 1 1.29.12 4.56 4.56 0 0 1 .92.22l-.3 1.33a1.83 1.83 0 0 0-.75-.4ZM287 50.81h8.32l-.12 2.79h-1l-.28-1.84-4-.14-.67 5 4.07-.14-.12 1.32-4-.14-.81 5.34 4.95-.21 1-2.12 1 .19-.71 3h-9.33l.07-.68 1.32-.21 1.4-11.19-1.19-.2ZM301.65 54.52a3.52 3.52 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.89 2.89 0 0 1 .53 1.94 16.85 16.85 0 0 1-.39 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.07.47.23.47a1.41 1.41 0 0 0 .82-.44l.14.49a1.76 1.76 0 0 1-.37.4 3.17 3.17 0 0 1-.69.45 2 2 0 0 1-.9.2A1 1 0 0 1 303 63a14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.64.21-1.19.28-1.66a9.18 9.18 0 0 0 .1-1.33 2 2 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33h-1.67l1.13-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.56 6.56 0 0 1 1.68-1.71ZM313.74 66.64a4.23 4.23 0 0 1-1.77 1.2 5.86 5.86 0 0 1-2 .37 7.37 7.37 0 0 1-1.13-.09 4.89 4.89 0 0 1-.87-.22l.37-1.42a6.44 6.44 0 0 0 1 .39 4.29 4.29 0 0 0 1.22.18 2.62 2.62 0 0 0 1.66-.43 2.58 2.58 0 0 0 .71-1.62l.33-2.33a4.25 4.25 0 0 1-1.18 1 3 3 0 0 1-1.52.39 2.52 2.52 0 0 1-2.17-1 4.5 4.5 0 0 1-.72-2.68 7.33 7.33 0 0 1 .66-3 6.45 6.45 0 0 1 1.95-2.49 4.73 4.73 0 0 1 3-1 5.1 5.1 0 0 1 1 .11 5.19 5.19 0 0 1 .91.28l1.09-.39-1.58 10.63a3.73 3.73 0 0 1-.96 2.12ZM313 54.81a2.66 2.66 0 0 0-2 .85 5.43 5.43 0 0 0-1.22 2.11 7.91 7.91 0 0 0-.4 2.46 3.81 3.81 0 0 0 .42 2 1.31 1.31 0 0 0 1.17.71 2.42 2.42 0 0 0 1.37-.44 3.83 3.83 0 0 0 1-1l.93-6.28a2.13 2.13 0 0 0-1.27-.41ZM320.16 62.51a.56.56 0 0 0 .08.32.24.24 0 0 0 .2.12.8.8 0 0 0 .4-.12 4.16 4.16 0 0 0 .5-.41l.23.42a3.57 3.57 0 0 1-.78.78 2 2 0 0 1-1.24.4 1.32 1.32 0 0 1-.89-.27.92.92 0 0 1-.25-.82c0-.29.18-1.43.46-3.4.16-1.07.36-2.48.59-4.23l-1.09-.4.11-.6 2.49-.4.39.19-1.18 8.25Zm-.08-10.81a.93.93 0 0 1-.27-.68 1.25 1.25 0 0 1 .35-.89 1 1 0 0 1 .81-.37.93.93 0 0 1 .7.27.9.9 0 0 1 .27.66 1.36 1.36 0 0 1-.34.92 1.11 1.11 0 0 1-.88.37.87.87 0 0 1-.64-.28ZM328.56 54.52a3.54 3.54 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.6 2.89 2.89 0 0 1 .53 1.94 16.85 16.85 0 0 1-.39 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.08.47.23.47a1.41 1.41 0 0 0 .82-.44l.23.49a1.76 1.76 0 0 1-.37.4 3.17 3.17 0 0 1-.69.45 2 2 0 0 1-.9.2 1 1 0 0 1-1.16-1.05 14.6 14.6 0 0 1 .32-2.14l.16-.84.07-.42c.11-.64.21-1.19.28-1.66a9.18 9.18 0 0 0 .1-1.33 2 2 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33h-1.67l1.13-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.56 6.56 0 0 1 1.71-1.71ZM340.41 63.44a4.26 4.26 0 0 1-2.23.6 3.35 3.35 0 0 1-1.95-.54 3 3 0 0 1-1.15-1.4 4.66 4.66 0 0 1-.37-1.87 7.38 7.38 0 0 1 .67-3.14 5.91 5.91 0 0 1 1.86-2.33 4.29 4.29 0 0 1 2.62-.88 2.62 2.62 0 0 1 1.88.6 2 2 0 0 1 .63 1.56 2.63 2.63 0 0 1-1 2.1 6.47 6.47 0 0 1-2.37 1.28 9.38 9.38 0 0 1-2.46.46A4 4 0 0 0 337 62a1.69 1.69 0 0 0 1.6.9 3.61 3.61 0 0 0 1.54-.36 4.27 4.27 0 0 0 1.39-1l.39.65a4.68 4.68 0 0 1-1.51 1.25Zm-2.4-8a4.53 4.53 0 0 0-1 1.65 6.69 6.69 0 0 0-.44 1.94 6.13 6.13 0 0 0 1.85-.39 4.25 4.25 0 0 0 1.6-1 2 2 0 0 0 .65-1.46 1.55 1.55 0 0 0-.32-1.05 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.46.68ZM349.39 63.44a4.26 4.26 0 0 1-2.23.6 3.35 3.35 0 0 1-2-.54 3.09 3.09 0 0 1-1.15-1.4 4.66 4.66 0 0 1-.37-1.87 7.38 7.38 0 0 1 .68-3.14 5.91 5.91 0 0 1 1.86-2.33 4.25 4.25 0 0 1 2.62-.88 2.65 2.65 0 0 1 1.88.6 2 2 0 0 1 .63 1.56 2.66 2.66 0 0 1-1 2.1 6.51 6.51 0 0 1-2.31 1.28 9.38 9.38 0 0 1-2.46.46 3.94 3.94 0 0 0 .43 2.13 1.66 1.66 0 0 0 1.59.9 3.65 3.65 0 0 0 1.55-.36 4.34 4.34 0 0 0 1.38-1l.39.65a4.68 4.68 0 0 1-1.49 1.24Zm-2.4-8a4.67 4.67 0 0 0-1 1.65 7 7 0 0 0-.44 1.94 6.07 6.07 0 0 0 1.85-.39 4.38 4.38 0 0 0 1.61-1 2 2 0 0 0 .65-1.46 1.5 1.5 0 0 0-.33-1.05 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.44.68ZM356.34 55.24a4.21 4.21 0 0 1 1.13-1 2.81 2.81 0 0 1 1.4-.38.93.93 0 0 1 .56.12l-.37 1.86a1.64 1.64 0 0 0-.86-.18 2.68 2.68 0 0 0-1.48.5 3.31 3.31 0 0 0-1.14 1.4l-.8 6.24h-1.69l1.11-8.45-.88-.44.12-.65 2.07-.4.35.21-.12 1.56-.08.7a3.9 3.9 0 0 1 .68-1.09ZM362.32 62.51a.56.56 0 0 0 .08.32.23.23 0 0 0 .2.12.8.8 0 0 0 .4-.12 4.16 4.16 0 0 0 .5-.41l.22.42a3.18 3.18 0 0 1-.78.78 2 2 0 0 1-1.23.4 1.3 1.3 0 0 1-.89-.27.92.92 0 0 1-.25-.82c0-.29.17-1.43.45-3.4.17-1.07.36-2.48.6-4.23l-1.09-.4.11-.6 2.49-.4.38.19-1.17 8.25Zm-.08-10.81a.93.93 0 0 1-.24-.7 1.24 1.24 0 0 1 .34-.89 1.07 1.07 0 0 1 .82-.37.94.94 0 0 1 .7.27.93.93 0 0 1 .26.66 1.35 1.35 0 0 1-.33.92 1.12 1.12 0 0 1-.88.37.86.86 0 0 1-.67-.26ZM370.36 54.52a3.55 3.55 0 0 1 2-.64 1.91 1.91 0 0 1 1.51.6 2.89 2.89 0 0 1 .52 1.94 17.7 17.7 0 0 1-.38 2.69l-.07.4-.14.84a12.68 12.68 0 0 0-.28 2.13c0 .31.07.47.22.47a1.4 1.4 0 0 0 .83-.44l.21.46a1.76 1.76 0 0 1-.37.4 3.74 3.74 0 0 1-.69.45 2.13 2.13 0 0 1-.91.2 1 1 0 0 1-1.1-1.07 14.1 14.1 0 0 1 .31-2.14l.16-.84.07-.42c.12-.64.21-1.19.28-1.66a9.27 9.27 0 0 0 .11-1.33 1.88 1.88 0 0 0-.24-1.11 1 1 0 0 0-.83-.31 2.59 2.59 0 0 0-1.57.7 6.6 6.6 0 0 0-1.48 1.67l-.86 6.33H366l1.12-8.49-.86-.42.14-.66 2.14-.39.37.19-.23 2.11a6.72 6.72 0 0 1 1.68-1.66ZM382.46 66.64a4.23 4.23 0 0 1-1.77 1.2 5.87 5.87 0 0 1-2 .37 7.34 7.34 0 0 1-1.12-.09 4.82 4.82 0 0 1-.88-.22l.37-1.42a6.58 6.58 0 0 0 1 .39 4.24 4.24 0 0 0 1.22.18 2.57 2.57 0 0 0 1.65-.43 2.53 2.53 0 0 0 .72-1.62l.33-2.33a4.25 4.25 0 0 1-1.18 1 3.07 3.07 0 0 1-1.52.39 2.52 2.52 0 0 1-2.17-1 4.5 4.5 0 0 1-.73-2.68 7.35 7.35 0 0 1 .67-3 6.27 6.27 0 0 1 1.95-2.51 4.71 4.71 0 0 1 3-1 5.1 5.1 0 0 1 1 .11 5.47 5.47 0 0 1 .91.28l1.09-.39-1.58 10.63a3.86 3.86 0 0 1-.96 2.14Zm-.7-11.83a2.62 2.62 0 0 0-2 .85 5.31 5.31 0 0 0-1.22 2.11 7.93 7.93 0 0 0-.41 2.46 3.7 3.7 0 0 0 .43 2 1.31 1.31 0 0 0 1.17.71 2.42 2.42 0 0 0 1.37-.44 3.79 3.79 0 0 0 1-1l.93-6.28a2.08 2.08 0 0 0-1.27-.41ZM295.85 80.3q1.44-3.12 1.89-4.29h3.76l-.07.75-1.56.19-.34 11.2 1.49.21-.02.64h-4.82l.07-.68 1.6-.21.4-7.76.23-3.56L297.1 80l-4.1 9.08-.94.12-1.44-9-.55-3.33-.71 3.4-1.88 7.9 1.77.21-.07.68h-4.53l.07-.68 1.46-.21L289 77l-1.62-.19.11-.75h4.07l1.35 8.56.23 1.79.77-1.76q.49-1.22 1.94-4.35ZM308 88.64a4.35 4.35 0 0 1-2.23.59 3.38 3.38 0 0 1-2-.53 3.26 3.26 0 0 1-1.15-1.4 4.87 4.87 0 0 1-.36-1.87 7.34 7.34 0 0 1 .67-3.14 6 6 0 0 1 1.91-2.29 4.32 4.32 0 0 1 2.62-.87 2.63 2.63 0 0 1 1.88.59 2.06 2.06 0 0 1 .63 1.56 2.65 2.65 0 0 1-1 2.11 6.62 6.62 0 0 1-2.37 1.28 10 10 0 0 1-2.46.46 3.89 3.89 0 0 0 .42 2.13 1.69 1.69 0 0 0 1.6.9 3.61 3.61 0 0 0 1.54-.36 4.3 4.3 0 0 0 1.39-1l.38.65a4.52 4.52 0 0 1-1.47 1.19Zm-2.41-8a4.68 4.68 0 0 0-1 1.64 7.5 7.5 0 0 0-.44 1.94 6.11 6.11 0 0 0 1.86-.38 4.61 4.61 0 0 0 1.6-1 2 2 0 0 0 .65-1.47 1.54 1.54 0 0 0-.32-1 1.15 1.15 0 0 0-.89-.36 2 2 0 0 0-1.45.64ZM318.78 87.71c0 .29.09.44.26.44a1.42 1.42 0 0 0 .81-.44l.24.44a3.31 3.31 0 0 1-.76.71 1.9 1.9 0 0 1-1.15.36c-.69 0-1-.38-1-1.13v-.38a4.57 4.57 0 0 1-1.23 1.07 3.18 3.18 0 0 1-1.65.44 2.46 2.46 0 0 1-2.14-1 4.54 4.54 0 0 1-.72-2.68 7.24 7.24 0 0 1 .67-3 6.3 6.3 0 0 1 1.89-2.48 4.67 4.67 0 0 1 3-1 4.42 4.42 0 0 1 1.31.21l.48-3.16-1.16-.3.14-.61 2.47-.32.39.23-1.83 12.4a.59.59 0 0 0-.02.2Zm-2-7.7a2.62 2.62 0 0 0-2 .85 5.14 5.14 0 0 0-1.21 2.1 8 8 0 0 0-.4 2.47 3.83 3.83 0 0 0 .42 2 1.27 1.27 0 0 0 1.16.7 2.39 2.39 0 0 0 1.4-.46 3.7 3.7 0 0 0 1-1l.95-6.19a2.2 2.2 0 0 0-1.32-.48ZM324.44 87.71a.56.56 0 0 0 .08.32.25.25 0 0 0 .21.12.79.79 0 0 0 .39-.13 3.18 3.18 0 0 0 .5-.4l.23.42a3.35 3.35 0 0 1-.78.77 2 2 0 0 1-1.24.41 1.34 1.34 0 0 1-.89-.28.92.92 0 0 1-.25-.81c0-.29.18-1.43.46-3.41.16-1.06.36-2.47.59-4.22l-1.08-.41.1-.59 2.49-.41.39.2-1.18 8.24Zm-.07-10.81a1 1 0 0 1-.28-.68 1.29 1.29 0 0 1 .35-.9 1.1 1.1 0 0 1 .81-.37.92.92 0 0 1 1 .93 1.29 1.29 0 0 1-.34.92 1.11 1.11 0 0 1-.87.38.84.84 0 0 1-.67-.28ZM332.76 80.18a2.22 2.22 0 0 0-1.53.62 4.15 4.15 0 0 0-1.06 1.74 8 8 0 0 0-.37 2.55 4.09 4.09 0 0 0 .49 2.24 1.59 1.59 0 0 0 1.42.75 2.71 2.71 0 0 0 1.29-.28 5.56 5.56 0 0 0 1.15-.81l.35.65a4.84 4.84 0 0 1-1.3 1.08 3.94 3.94 0 0 1-2 .5 2.78 2.78 0 0 1-2.31-1 4.49 4.49 0 0 1-.83-2.9 7.35 7.35 0 0 1 .6-2.87 6.1 6.1 0 0 1 1.72-2.45 4.26 4.26 0 0 1 2.77-.94 5.66 5.66 0 0 1 1.06.1 3.31 3.31 0 0 1 .92.3l-.47 1.7-.3-.1a2.88 2.88 0 0 0-1.6-.88ZM338.88 87.71a.56.56 0 0 0 .12.29.23.23 0 0 0 .2.12.82.82 0 0 0 .4-.13 3.18 3.18 0 0 0 .5-.4l.23.42a3.56 3.56 0 0 1-.78.77 2 2 0 0 1-1.24.41 1.32 1.32 0 0 1-.89-.28.9.9 0 0 1-.25-.81c0-.29.18-1.43.46-3.41.16-1.06.36-2.47.59-4.22l-1.09-.41.11-.59 2.49-.41.39.2-1.18 8.24Zm-.08-10.81a1 1 0 0 1-.27-.68 1.29 1.29 0 0 1 .35-.9 1.07 1.07 0 0 1 .81-.37.92.92 0 0 1 1 .93 1.34 1.34 0 0 1-.34.92 1.11 1.11 0 0 1-.88.38.87.87 0 0 1-.67-.28ZM347.19 79.72a3.54 3.54 0 0 1 2-.64 1.89 1.89 0 0 1 1.51.59 2.93 2.93 0 0 1 .53 2 16.77 16.77 0 0 1-.39 2.68l-.07.41-.14.84a12.6 12.6 0 0 0-.28 2.12c0 .32.08.48.23.48a1.38 1.38 0 0 0 .82-.44l.21.45a1.91 1.91 0 0 1-.36.41 3.11 3.11 0 0 1-.7.44 1.93 1.93 0 0 1-.9.21 1 1 0 0 1-1.11-1.07 14.24 14.24 0 0 1 .32-2.14l.16-.85.07-.42c.11-.63.21-1.18.28-1.66a9 9 0 0 0 .1-1.32 1.87 1.87 0 0 0-.23-1.11 1 1 0 0 0-.84-.31 2.66 2.66 0 0 0-1.57.69 6.48 6.48 0 0 0-1.48 1.68l-.89 6.24h-1.66l1.12-8.49-.86-.42.14-.67 2.14-.38.37.19-.23 2.1a6.52 6.52 0 0 1 1.71-1.61ZM359 88.64a4.35 4.35 0 0 1-2.23.59 3.34 3.34 0 0 1-1.95-.53 3.19 3.19 0 0 1-1.15-1.4 4.69 4.69 0 0 1-.37-1.87 7.34 7.34 0 0 1 .67-3.14 6 6 0 0 1 1.9-2.29 4.35 4.35 0 0 1 2.63-.87 2.61 2.61 0 0 1 1.87.59 2 2 0 0 1 .64 1.56 2.68 2.68 0 0 1-1 2.11 6.75 6.75 0 0 1-2.38 1.28 9.81 9.81 0 0 1-2.46.46 4.08 4.08 0 0 0 .42 2.13 1.69 1.69 0 0 0 1.6.9 3.65 3.65 0 0 0 1.55-.36 4.37 4.37 0 0 0 1.38-1l.39.65a4.57 4.57 0 0 1-1.51 1.19Zm-2.4-8a4.54 4.54 0 0 0-1 1.64 6.76 6.76 0 0 0-.44 1.94 6 6 0 0 0 1.85-.38 4.52 4.52 0 0 0 1.6-1 2 2 0 0 0 .65-1.47 1.49 1.49 0 0 0-.32-1 1.14 1.14 0 0 0-.89-.36 2 2 0 0 0-1.41.64Z" class="cls-2" transform="translate(-24.97 -24.05)"/> </svg> </a> </div> <div class="global-nav__controls"><button class="controls__menu"> Close <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M20.728 8l1.818 1.818-5.455 5.455 5.455 5.455-1.818 1.818-5.455-5.455-5.455 5.455L8 20.728l5.455-5.455L8 9.818 9.818 8l5.455 5.455z"></path></svg> </button> </div></div> <div class="mega-menu__content"><div> <button class="accordion">About Us</button> <div class="panel"><ul><li><a href="https://www.nationalacademies.org/about"> About Us </a></li><li><a href="https://www.nationalacademies.org/about/organization"> Our Program Divisions </a></li><li><a href="https://www.nationalacademies.org/about/the-national-academies"> Our Three Academies </a></li><li><a href="https://www.nationalacademies.org/about/leadership"> Leadership </a></li><li><a href="https://www.nationalacademies.org/newsroom"> Newsroom </a></li><li><a href="https://www.nationalacademies.org/ocga"> Government Affairs </a></li><li><a href="https://www.nationalacademies.org/giving"> Giving </a></li><li><a href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/statement-on-diversity-and-inclusion"> Statement on Diversity and Inclusion </a></li></ul> </div> <button class="no-accordion"><a href="https://www.nationalacademies.org/events">Events</a></button> <button class="accordion">Our Work</button> <div class="panel"><ul><li><a href="https://www.nationalacademies.org/our-work"> Our Work </a></li><li><a href="https://www.nationalacademies.org/about/our-study-process"> Our Study Process </a></li><li><a href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/conflict-of-interest-policies-and-procedures"> Conflict of Interest Policies and Procedures </a></li><li><a href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/project-comments-and-information"> Project Comments and Information </a></li></ul> </div> <button class="accordion">Publications</button> <div class="panel"> <ul><li><a href="https://nap.nationalacademies.org/" target="_self"> Read Our Expert Reports and Published Proceedings </a></li><li><a href="https://www.pnas.org/" target="_self"> Explore PNAS, the Flagship Scientific Journal of NAS </a></li><li><a href="http://pubsindex.trb.org" target="_self"> Access Transportation Research Board Publications </a></li></ul> </div> <button class="accordion">Topics</button> <div class="panel"> <ul><li><a href="https://www.nationalacademies.org/topics/covid-19-resources"> Coronavirus Disease 2019 (COVID-19) </a></li><li><a href="https://www.nationalacademies.org/topics/resources-on-diversity-equity-and-inclusion"> Diversity, Equity, and Inclusion </a></li><li><a href="https://www.nationalacademies.org/topics/climate"> Climate </a></li><li><a href="https://www.nationalacademies.org/topics/economic-recovery"> Economic Recovery </a></li></ul> </div> <button class="no-accordion"><a href="https://www.nationalacademies.org/engagement">Engagement</a></button> <button class="accordion">Opportunities</button> <div class="panel"> <ul><li><a href="https://www.nationalacademies.org/careers"> Careers </a></li><li><a href="https://www.nationalacademies.org/fellowships-and-grants"> Fellowships and Grants </a></li><li><a href="https://www.nationalacademies.org/about/volunteers"> Volunteers </a></li><li><a href="https://www.nationalacademies.org/about/sponsors"> Sponsors </a></li><li><a href="https://www.nationalacademies.org/about/members"> Members </a></li></ul> </div> </div></div> </div> </div> <!--Mobile menu end --> </div> <div class="container flex justify-between"> <div class="global-nav__logo" > <a class="navbar-brand js-scroll-trigger" aria-label="Logo of The National Academies" href="https://www.nationalacademies.org/home"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 315.11 58.33"> <defs><style>.cls-2{fill:#211747;}</style></defs><g id="Layer_2" data-name="Layer 2"><g id="Layer_1-2" data-name="Layer 1"><polygon class="cls-2" points="3.73 7.55 20.84 25.49 20.84 1.5 17.81 1.5 17.81 18.08 0.7 0.14 0.7 24.06 3.73 24.06 3.73 7.55"/><polygon class="cls-2" points="28.49 24.06 31.84 24.06 38.98 7.25 45.92 23.87 45.99 24.06 49.34 24.06 39.05 0 28.49 24.06"/><polygon class="cls-2" points="153.23 24.06 156.58 24.06 163.72 7.25 170.66 23.87 170.73 24.06 174.08 24.06 163.79 0 153.23 24.06"/><polygon class="cls-2" points="59.72 24.06 62.76 24.06 62.76 4.41 69.52 4.41 69.52 1.5 52.96 1.5 52.96 4.41 59.72 4.41 59.72 24.06"/><rect class="cls-2" x="78.85" y="1.5" width="3.03" height="22.55"/><path class="cls-2" d="M115.69,12.8a11.72,11.72,0,0,0-23.44,0,11.72,11.72,0,0,0,23.44,0m-3.09,0A8.63,8.63,0,1,1,104,4a8.7,8.7,0,0,1,8.63,8.76"/><polygon class="cls-2" points="128.29 7.55 145.4 25.49 145.4 1.5 142.37 1.5 142.37 18.08 125.25 0.14 125.25 24.06 128.29 24.06 128.29 7.55"/><polygon class="cls-2" points="184.65 1.5 181.62 1.5 181.62 24.06 193.7 24.06 193.7 21.16 184.65 21.16 184.65 1.5"/><path class="cls-2" d="M37.15,34.08A11.48,11.48,0,0,0,25.37,45.8,11.36,11.36,0,0,0,37,57.38a11.77,11.77,0,0,0,6.79-2.19l.13-.09V51.2l-.51.51A9.77,9.77,0,0,1,41,53.5a8.9,8.9,0,0,1-3.95,1A8.75,8.75,0,0,1,37,37a9.09,9.09,0,0,1,3.66.82,10.71,10.71,0,0,1,2.78,2l.51.5V36.4l-.12-.09a10.62,10.62,0,0,0-6.67-2.23"/><path class="cls-2" d="M91.91,37.08a12.19,12.19,0,0,0-8.29-2.64H78.85V57h4.74c3.72,0,6-.73,8.22-2.61a10.75,10.75,0,0,0,4-8.65,11,11,0,0,0-3.86-8.65m-2.09,15.1a9.11,9.11,0,0,1-6.1,1.91H81.88V37.34h1.84c5.44,0,9,3.29,9,8.39a8.09,8.09,0,0,1-2.85,6.45"/><polygon class="cls-2" points="104.99 56.99 118.4 56.99 118.4 54.09 108.02 54.09 108.02 46.16 117.11 46.16 117.11 43.26 108.02 43.26 108.02 37.34 118.4 37.34 118.4 34.44 104.99 34.44 104.99 56.99"/><polygon class="cls-2" points="176.17 56.99 189.58 56.99 189.58 54.09 179.2 54.09 179.2 46.16 188.28 46.16 188.28 43.26 179.2 43.26 179.2 37.34 189.58 37.34 189.58 34.44 176.17 34.44 176.17 56.99"/><polygon class="cls-2" points="138.88 51.34 130.31 32.49 126.44 56.99 129.55 56.99 131.62 42.47 138.84 58.33 146.15 42.47 148.13 56.99 151.23 56.99 147.61 32.45 138.88 51.34"/><rect class="cls-2" x="161.31" y="34.44" width="3.03" height="22.55"/><path class="cls-2" d="M209,46.15A10.78,10.78,0,0,0,205.49,44l-2.12-.92a6.8,6.8,0,0,1-2.21-1.33,2,2,0,0,1-.73-1.69A3.07,3.07,0,0,1,203.82,37a3.62,3.62,0,0,1,2.28.69,4.87,4.87,0,0,1,1.2,1.55l.15.3,2.45-1.62-.12-.24a7,7,0,0,0-2.24-2.55,6.34,6.34,0,0,0-3.69-1c-3.8,0-6.45,2.45-6.45,6a4.86,4.86,0,0,0,1.6,3.8,9.51,9.51,0,0,0,2.84,1.78l2.15,1a9.5,9.5,0,0,1,2.56,1.46,3,3,0,0,1,1,2.35,4.2,4.2,0,0,1-7.06,2.92,4.93,4.93,0,0,1-1.24-2.64l-.05-.33-3,.82.06.28a8.32,8.32,0,0,0,2.13,4,6.74,6.74,0,0,0,5,1.87,6.91,6.91,0,0,0,7.24-7A5.58,5.58,0,0,0,209,46.15"/><polygon class="cls-2" points="0 56.99 3.35 56.99 10.49 40.18 17.43 56.81 17.5 56.99 20.86 56.99 10.56 32.93 0 56.99"/><polygon class="cls-2" points="50.21 56.99 53.57 56.99 60.71 40.18 67.64 56.81 67.72 56.99 71.07 56.99 60.78 32.93 50.21 56.99"/><path class="cls-2" d="M233.74,2.1a3.34,3.34,0,0,0-.93-.11,2.53,2.53,0,0,0-1.52.45,1.48,1.48,0,0,0-.63,1.28,2,2,0,0,0,.54,1.44,7.94,7.94,0,0,0,1.64,1.26,13.5,13.5,0,0,1,1.26.89,3.76,3.76,0,0,1,.83,1,2.62,2.62,0,0,1,.33,1.33,3.06,3.06,0,0,1-.5,1.74,3.26,3.26,0,0,1-1.4,1.18,5.06,5.06,0,0,1-2.09.41,8.3,8.3,0,0,1-1.94-.22,3.51,3.51,0,0,1-1.24-.46l.34-2.18h.92l.09,1.62a2.53,2.53,0,0,0,.81.33,3.86,3.86,0,0,0,1,.13,2.93,2.93,0,0,0,1.77-.53,1.68,1.68,0,0,0,.73-1.47,1.91,1.91,0,0,0-.49-1.34,6.42,6.42,0,0,0-1.49-1.16c-.52-.34-1-.65-1.32-.94a3.76,3.76,0,0,1-.9-1,2.64,2.64,0,0,1-.37-1.4,2.64,2.64,0,0,1,.5-1.62,3.23,3.23,0,0,1,1.35-1.08,4.58,4.58,0,0,1,1.86-.37,7,7,0,0,1,1.72.18,3.31,3.31,0,0,1,1.09.45l-.35,2h-.83l-.12-1.47a2.39,2.39,0,0,0-.68-.3"/><path class="cls-2" d="M241,5a2,2,0,0,0-1.35.54,3.66,3.66,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,1.95,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a3.84,3.84,0,0,1-1.14.95,3.43,3.43,0,0,1-1.76.43,2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54A6.43,6.43,0,0,1,237.34,7a5.3,5.3,0,0,1,1.55-2.08,3.74,3.74,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3.09,3.09,0,0,1,.81.26l-.42,1.49-.26-.09A2.53,2.53,0,0,0,241,5"/><path class="cls-2" d="M246.1,11.61a.56.56,0,0,0,.07.29.21.21,0,0,0,.18.1.65.65,0,0,0,.34-.11,2.64,2.64,0,0,0,.44-.35l.2.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.31-2.16.52-3.7l-1-.35.09-.52,2.18-.36.34.17-1,7.22ZM246,2.15a.84.84,0,0,1-.23-.59,1.13,1.13,0,0,1,.3-.79.94.94,0,0,1,.71-.32.79.79,0,0,1,.61.24.77.77,0,0,1,.23.57,1.15,1.15,0,0,1-.29.81,1,1,0,0,1-.77.33.75.75,0,0,1-.56-.25"/><path class="cls-2" d="M254,12.43a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.72-.47,2.74,2.74,0,0,1-1-1.23A4,4,0,0,1,249,9.62a6.35,6.35,0,0,1,.59-2.75,5.08,5.08,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.27,2.27,0,0,1,1.64.52A1.79,1.79,0,0,1,255.72,6a2.34,2.34,0,0,1-.89,1.84,5.79,5.79,0,0,1-2.08,1.12,8.1,8.1,0,0,1-2.15.4,3.42,3.42,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57A4,4,0,0,1,254,12.43m-2.11-7A4.1,4.1,0,0,0,251,6.88a5.9,5.9,0,0,0-.39,1.69,5.31,5.31,0,0,0,1.62-.33,4,4,0,0,0,1.41-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-2" d="M261.24,4.62A3.14,3.14,0,0,1,263,4.06a1.67,1.67,0,0,1,1.32.52,2.61,2.61,0,0,1,.46,1.71,15.44,15.44,0,0,1-.34,2.34l-.06.36-.13.74a11.51,11.51,0,0,0-.24,1.85c0,.28.06.42.2.42a1.22,1.22,0,0,0,.72-.39l.18.4a1.54,1.54,0,0,1-.32.36,2.92,2.92,0,0,1-.61.39,1.73,1.73,0,0,1-.79.17.84.84,0,0,1-1-.93,12.88,12.88,0,0,1,.27-1.88l.14-.73.06-.37c.1-.55.19-1,.25-1.45a8.17,8.17,0,0,0,.09-1.16,1.67,1.67,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.33,2.33,0,0,0-1.37.61,5.75,5.75,0,0,0-1.3,1.47l-.75,5.54h-1.46l1-7.43L257.62,5l.12-.58,1.88-.34.32.17-.2,1.84a5.72,5.72,0,0,1,1.5-1.45"/><path class="cls-2" d="M270.77,5a2,2,0,0,0-1.35.54,3.66,3.66,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,1.95,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a3.84,3.84,0,0,1-1.14.95,3.43,3.43,0,0,1-1.76.43,2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54A6.43,6.43,0,0,1,267.13,7a5.3,5.3,0,0,1,1.55-2.08,3.76,3.76,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3,3,0,0,1,.81.26l-.42,1.49-.26-.09A2.53,2.53,0,0,0,270.77,5"/><path class="cls-2" d="M278.66,12.43a3.75,3.75,0,0,1-2,.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63,6.35,6.35,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52A1.82,1.82,0,0,1,280.38,6a2.34,2.34,0,0,1-.89,1.84,5.79,5.79,0,0,1-2.08,1.12,8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87A1.48,1.48,0,0,0,277,12a3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-2" d="M286.18,5.13a2.75,2.75,0,0,0-1.09-.21,1.85,1.85,0,0,0-1.13.31,1,1,0,0,0-.43.89,1.43,1.43,0,0,0,.34.93,6.08,6.08,0,0,0,1,.88,10.43,10.43,0,0,1,.88.74,3.1,3.1,0,0,1,.6.76,2.15,2.15,0,0,1,.26,1.05,2.09,2.09,0,0,1-.9,1.81,3.71,3.71,0,0,1-2.22.64,6.32,6.32,0,0,1-1.37-.15,2.1,2.1,0,0,1-.9-.34l.24-1.06a3.41,3.41,0,0,0,.88.44,3.37,3.37,0,0,0,1.18.21,2.21,2.21,0,0,0,1.15-.29,1,1,0,0,0,.48-.92,1.37,1.37,0,0,0-.38-1,6.83,6.83,0,0,0-1.1-.89,3.07,3.07,0,0,1-1.58-2.5,2.16,2.16,0,0,1,.4-1.27,2.64,2.64,0,0,1,1.09-.84,3.81,3.81,0,0,1,1.54-.3,6.32,6.32,0,0,1,1.13.1,4.07,4.07,0,0,1,.8.19l-.26,1.17a1.55,1.55,0,0,0-.6-.39"/><polygon class="cls-2" points="229.34 23.43 236.61 23.43 236.51 25.87 235.6 25.87 235.35 24.25 231.87 24.13 231.29 28.48 234.85 28.36 234.74 29.51 231.21 29.38 230.58 34.05 234.91 33.87 235.77 32.01 236.63 32.18 236.01 34.84 227.85 34.84 227.91 34.24 229.06 34.05 230.29 24.25 229.24 24.09 229.34 23.43"/><path class="cls-2" d="M242.18,26.67a3.16,3.16,0,0,1,1.73-.56,1.67,1.67,0,0,1,1.32.53,2.51,2.51,0,0,1,.46,1.7,14.42,14.42,0,0,1-.34,2.35l-.06.35-.12.74a10.81,10.81,0,0,0-.25,1.86c0,.27.07.41.2.41a1.15,1.15,0,0,0,.72-.38l.19.4a1.54,1.54,0,0,1-.32.35,2.46,2.46,0,0,1-.61.39,1.76,1.76,0,0,1-.79.18c-.65,0-1-.31-1-.94a11.94,11.94,0,0,1,.28-1.87l.14-.74.06-.37c.1-.55.18-1,.24-1.45a8.14,8.14,0,0,0,.1-1.16,1.64,1.64,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.3,2.3,0,0,0-1.38.61,5.67,5.67,0,0,0-1.29,1.46l-.76,5.55h-1.45l1-7.44-.76-.36.13-.59,1.87-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-2" d="M252.77,37.29a4,4,0,0,1-1.55,1.05,5.36,5.36,0,0,1-1.78.32,5.69,5.69,0,0,1-1-.09,3.45,3.45,0,0,1-.76-.19l.32-1.24a5,5,0,0,0,.9.34,3.56,3.56,0,0,0,1.06.17,2.22,2.22,0,0,0,1.45-.39,2.12,2.12,0,0,0,.63-1.41l.29-2a3.73,3.73,0,0,1-1,.84A2.61,2.61,0,0,1,250,35a2.16,2.16,0,0,1-1.89-.88,3.9,3.9,0,0,1-.64-2.35,6.52,6.52,0,0,1,.58-2.6A5.62,5.62,0,0,1,249.73,27a4.15,4.15,0,0,1,2.61-.87,4.47,4.47,0,0,1,.87.1,5.15,5.15,0,0,1,.8.24l1-.34-1.38,9.31a3.37,3.37,0,0,1-.81,1.87m-.61-10.36a2.36,2.36,0,0,0-1.75.74,4.82,4.82,0,0,0-1.07,1.85,7.16,7.16,0,0,0-.35,2.15,3.3,3.3,0,0,0,.37,1.72,1.16,1.16,0,0,0,1,.62,2.09,2.09,0,0,0,1.2-.39,3.12,3.12,0,0,0,.9-.87l.82-5.5a1.85,1.85,0,0,0-1.15-.32"/><path class="cls-2" d="M258.39,33.67a.45.45,0,0,0,.07.28.19.19,0,0,0,.17.1.6.6,0,0,0,.35-.11,3.13,3.13,0,0,0,.44-.35l.2.37a3.17,3.17,0,0,1-.69.68,1.77,1.77,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.24.82.82,0,0,1-.22-.71c0-.26.16-1.25.4-3,.15-.93.32-2.17.52-3.7l-.95-.36.09-.52,2.19-.35.33.17-1,7.21Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.09,1.09,0,0,1,.3-.78.94.94,0,0,1,.72-.33.81.81,0,0,1,.61.24.82.82,0,0,1,.23.58,1.17,1.17,0,0,1-.29.8,1,1,0,0,1-.77.33.76.76,0,0,1-.56-.24"/><path class="cls-2" d="M265.74,26.67a3.11,3.11,0,0,1,1.72-.56,1.65,1.65,0,0,1,1.32.53,2.5,2.5,0,0,1,.47,1.7,15.68,15.68,0,0,1-.34,2.35l-.06.35-.13.74a11.6,11.6,0,0,0-.24,1.86c0,.27.06.41.2.41a1.2,1.2,0,0,0,.72-.38l.18.4a1.54,1.54,0,0,1-.32.35,2.28,2.28,0,0,1-.61.39,1.73,1.73,0,0,1-.79.18.84.84,0,0,1-1-.94,12.8,12.8,0,0,1,.27-1.87l.14-.74.06-.37c.1-.55.19-1,.25-1.45a8.07,8.07,0,0,0,.09-1.16,1.64,1.64,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.27,2.27,0,0,0-1.37.61,5.7,5.7,0,0,0-1.3,1.46l-.75,5.55h-1.46l1-7.44-.75-.36.12-.59,1.88-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-2" d="M276.11,34.48a3.7,3.7,0,0,1-1.95.52,3,3,0,0,1-1.71-.46,2.87,2.87,0,0,1-1-1.23,4.24,4.24,0,0,1-.32-1.64,6.53,6.53,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.76,3.76,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.53,1.76,1.76,0,0,1,.55,1.36,2.31,2.31,0,0,1-.88,1.84A5.83,5.83,0,0,1,274.87,31a9.08,9.08,0,0,1-2.15.4,3.49,3.49,0,0,0,.36,1.86,1.48,1.48,0,0,0,1.4.79,3.16,3.16,0,0,0,1.35-.31,3.9,3.9,0,0,0,1.22-.9l.33.57a3.93,3.93,0,0,1-1.27,1.1m-2.1-7a4.12,4.12,0,0,0-.88,1.44,6.37,6.37,0,0,0-.38,1.7,5.17,5.17,0,0,0,1.62-.34,3.78,3.78,0,0,0,1.4-.87,1.74,1.74,0,0,0,.57-1.28,1.31,1.31,0,0,0-.28-.91,1,1,0,0,0-.78-.32,1.68,1.68,0,0,0-1.27.58"/><path class="cls-2" d="M284,34.48a3.7,3.7,0,0,1-2,.52,3,3,0,0,1-1.71-.46,2.87,2.87,0,0,1-1-1.23,4.24,4.24,0,0,1-.32-1.64,6.53,6.53,0,0,1,.59-2.75,5.26,5.26,0,0,1,1.63-2,3.79,3.79,0,0,1,2.3-.77,2.31,2.31,0,0,1,1.64.53,1.76,1.76,0,0,1,.55,1.36,2.31,2.31,0,0,1-.88,1.84A5.74,5.74,0,0,1,282.73,31a9.08,9.08,0,0,1-2.15.4,3.4,3.4,0,0,0,.37,1.86,1.45,1.45,0,0,0,1.39.79,3.2,3.2,0,0,0,1.36-.31,3.86,3.86,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.28,1.1m-2.1-7a4.12,4.12,0,0,0-.88,1.44,6.37,6.37,0,0,0-.38,1.7,5.17,5.17,0,0,0,1.62-.34,3.7,3.7,0,0,0,1.4-.87,1.74,1.74,0,0,0,.57-1.28,1.31,1.31,0,0,0-.28-.91,1,1,0,0,0-.78-.32,1.7,1.7,0,0,0-1.27.58"/><path class="cls-2" d="M290.05,27.3a3.77,3.77,0,0,1,1-.83,2.34,2.34,0,0,1,1.22-.34.83.83,0,0,1,.5.11l-.33,1.62a1.45,1.45,0,0,0-.75-.15,2.3,2.3,0,0,0-1.29.44,2.89,2.89,0,0,0-1,1.22l-.7,5.47h-1.48l1-7.41-.77-.38.11-.57,1.81-.35.31.18-.11,1.37-.08.61a4.09,4.09,0,0,1,.6-1"/><path class="cls-2" d="M295.29,33.67a.45.45,0,0,0,.07.28.19.19,0,0,0,.17.1.6.6,0,0,0,.35-.11,3.13,3.13,0,0,0,.44-.35l.2.37a3.17,3.17,0,0,1-.69.68,1.77,1.77,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.24.82.82,0,0,1-.22-.71c0-.26.16-1.25.4-3,.15-.93.32-2.17.52-3.7l-.95-.36.09-.52,2.19-.35.33.17-1,7.21Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.09,1.09,0,0,1,.3-.78.94.94,0,0,1,.72-.33.81.81,0,0,1,.61.24.82.82,0,0,1,.23.58,1.17,1.17,0,0,1-.29.8,1,1,0,0,1-.77.33.76.76,0,0,1-.56-.24"/><path class="cls-2" d="M302.33,26.67a3.14,3.14,0,0,1,1.73-.56,1.67,1.67,0,0,1,1.32.53,2.56,2.56,0,0,1,.46,1.7,15.68,15.68,0,0,1-.34,2.35l-.06.35-.12.74a10.81,10.81,0,0,0-.25,1.86c0,.27.07.41.2.41a1.2,1.2,0,0,0,.72-.38l.19.4a1.64,1.64,0,0,1-.33.35,2.22,2.22,0,0,1-.6.39,1.76,1.76,0,0,1-.79.18c-.65,0-1-.31-1-.94a12.8,12.8,0,0,1,.27-1.87l.14-.74.06-.37c.11-.55.19-1,.25-1.45a8.07,8.07,0,0,0,.09-1.16,1.73,1.73,0,0,0-.2-1,.88.88,0,0,0-.73-.27,2.3,2.3,0,0,0-1.38.61,5.7,5.7,0,0,0-1.3,1.46l-.75,5.55h-1.46l1-7.44-.75-.36.13-.59,1.87-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-2" d="M312.92,37.29a4,4,0,0,1-1.55,1.05,5.36,5.36,0,0,1-1.78.32,5.69,5.69,0,0,1-1-.09,3.57,3.57,0,0,1-.77-.19l.33-1.24a5,5,0,0,0,.9.34,3.51,3.51,0,0,0,1.06.17,2.22,2.22,0,0,0,1.45-.39,2.11,2.11,0,0,0,.62-1.41l.3-2a3.73,3.73,0,0,1-1,.84,2.63,2.63,0,0,1-1.33.34,2.16,2.16,0,0,1-1.89-.88,3.9,3.9,0,0,1-.64-2.35,6.35,6.35,0,0,1,.58-2.6A5.62,5.62,0,0,1,309.88,27a4.11,4.11,0,0,1,2.61-.87,4.56,4.56,0,0,1,.87.1,5.15,5.15,0,0,1,.8.24l.95-.34-1.38,9.31a3.37,3.37,0,0,1-.81,1.87m-.62-10.36a2.35,2.35,0,0,0-1.74.74,4.7,4.7,0,0,0-1.07,1.85,7.16,7.16,0,0,0-.35,2.15,3.3,3.3,0,0,0,.37,1.72,1.14,1.14,0,0,0,1,.62,2,2,0,0,0,1.19-.39,3.3,3.3,0,0,0,.91-.87l.82-5.5a1.88,1.88,0,0,0-1.16-.32"/><path class="cls-2" d="M237.11,49.24c.83-1.82,1.39-3.07,1.65-3.76h3.29l-.06.66-1.37.17-.29,9.8,1.31.18-.07.6h-4.22l.06-.6,1.4-.18.35-6.79.2-3.12L238.2,49l-3.57,8-.83.11-1.26-7.89-.47-2.92-.63,3-1.64,6.91,1.55.18-.06.6h-4l.06-.6,1.28-.18,2.47-9.8-1.41-.17.09-.66h3.56L234.56,53l.2,1.57.67-1.53c.28-.69.84-1.95,1.68-3.77"/><path class="cls-2" d="M247.75,56.54a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63,6.35,6.35,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52,1.82,1.82,0,0,1,.55,1.37,2.34,2.34,0,0,1-.89,1.84A5.79,5.79,0,0,1,246.5,53a8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87A1.76,1.76,0,0,0,248,50.2a1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-2" d="M257.17,55.72q0,.39.24.39a1.19,1.19,0,0,0,.7-.39l.22.39a3.42,3.42,0,0,1-.67.62,1.68,1.68,0,0,1-1,.31c-.6,0-.9-.32-.9-1v-.34a3.94,3.94,0,0,1-1.08.94,2.72,2.72,0,0,1-1.44.38,2.13,2.13,0,0,1-1.87-.88,4,4,0,0,1-.63-2.34,6.41,6.41,0,0,1,.58-2.62A5.56,5.56,0,0,1,253,49a4.1,4.1,0,0,1,2.61-.86,4,4,0,0,1,1.15.18l.41-2.76-1-.26.13-.54,2.16-.28.34.2-1.6,10.84a.79.79,0,0,0,0,.17M255.42,49a2.32,2.32,0,0,0-1.74.75,4.64,4.64,0,0,0-1.06,1.84,7.15,7.15,0,0,0-.34,2.16,3.44,3.44,0,0,0,.36,1.72,1.12,1.12,0,0,0,1,.61,2.06,2.06,0,0,0,1.23-.41,3.11,3.11,0,0,0,.9-.91l.83-5.42a1.94,1.94,0,0,0-1.2-.34"/><path class="cls-2" d="M262.14,55.72a.57.57,0,0,0,.06.29.24.24,0,0,0,.18.1.71.71,0,0,0,.35-.11,3.84,3.84,0,0,0,.44-.35l.19.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.19,1.19,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.32-2.16.52-3.7l-.95-.35.09-.52,2.18-.36.34.17-1,7.22Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.12,1.12,0,0,1,.3-.78,1,1,0,0,1,.71-.32.83.83,0,0,1,.62.24.81.81,0,0,1,.23.57,1.2,1.2,0,0,1-.29.81,1,1,0,0,1-.77.33.73.73,0,0,1-.56-.25"/><path class="cls-2" d="M269.42,49.14a2,2,0,0,0-1.35.54,3.6,3.6,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,2,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a4,4,0,0,1-1.14,1A3.43,3.43,0,0,1,268,57a2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54,6.43,6.43,0,0,1,.53-2.51A5.3,5.3,0,0,1,267.33,49a3.69,3.69,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3.09,3.09,0,0,1,.81.26L271.07,50l-.26-.09a2.53,2.53,0,0,0-1.39-.78"/><path class="cls-2" d="M274.77,55.72a.56.56,0,0,0,.07.29.24.24,0,0,0,.18.1.71.71,0,0,0,.35-.11,3.15,3.15,0,0,0,.43-.35l.2.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.19,1.19,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.32-2.16.52-3.7l-1-.35.09-.52,2.18-.36.34.17-1,7.22Zm-.07-9.46a.85.85,0,0,1-.23-.6,1.12,1.12,0,0,1,.3-.78.94.94,0,0,1,.71-.32.79.79,0,0,1,.61.24.78.78,0,0,1,.24.57,1.16,1.16,0,0,1-.3.81,1,1,0,0,1-.76.33.75.75,0,0,1-.57-.25"/><path class="cls-2" d="M282,48.73a3.09,3.09,0,0,1,1.73-.56,1.65,1.65,0,0,1,1.32.52,2.54,2.54,0,0,1,.46,1.7,15.09,15.09,0,0,1-.33,2.35l-.07.36-.12.73a11.6,11.6,0,0,0-.24,1.86c0,.28.06.42.19.42a1.22,1.22,0,0,0,.73-.39l.18.4a1.77,1.77,0,0,1-.32.36,3.21,3.21,0,0,1-.61.39,1.86,1.86,0,0,1-.79.17c-.64,0-1-.31-1-.93a13.42,13.42,0,0,1,.28-1.88l.14-.73.06-.37c.1-.56.18-1,.25-1.45a8.17,8.17,0,0,0,.09-1.16,1.67,1.67,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.31,2.31,0,0,0-1.37.61,5.58,5.58,0,0,0-1.3,1.47l-.75,5.54H278.2l1-7.43-.75-.37.12-.58,1.87-.34.33.17-.2,1.84A5.85,5.85,0,0,1,282,48.73"/><path class="cls-2" d="M292.42,56.54a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63A6.35,6.35,0,0,1,288,51a5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52,1.82,1.82,0,0,1,.55,1.37,2.34,2.34,0,0,1-.89,1.84A5.79,5.79,0,0,1,291.17,53a8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/></g></g> </svg> </a> </div> <div class='global-nav-nap-menu' id="header-large"> <div class="dropdown" id="sitelinks-mobile"> <button class="btn btn-secondary dropdown-toggle" type="button" id="dropdownMenuButton" data-toggle="dropdown" aria-haspopup="true" aria-expanded="false"> <span class="hidden-xxs">Navigate to</span> </button> <div class="dropdown-menu" aria-labelledby="dropdownMenuButton"> <a class="dropdown-item" href="/content/help">Help</a> <a class="dropdown-item" href="/content/help-with-ordering">Ordering Information</a> <a class="dropdown-item" href="/new" >New Releases</a> <a class="dropdown-item" href="/author">Browse by Division</a> <a class="dropdown-item" href="/topic" class="site-links"> Browse by Topic</a> </div> </div> <ul class="user-links"> <li><a href="/"><span class="icon-menu" data-icon=""> </span> NAP Home</a></li> <li class="dropdown nap-mega-menu mega-dropdown"> <a href="#" class="dropdown-toggle" data-toggle="dropdown"><span class="icon-menu" data-icon=""> </span> Browse <span class="caret"></span></a> <ul class="dropdown-menu mega-dropdown-menu row" id="nap-mega-menu-inner"> <li class="d-flex" > <ul class="browse-col" > <li class="dropdown-header"><a href="/author">Publications by Division</a></li> <li><a href="/author/DBASSE/">Division of Behavioral and Social Sciences and Education</a> </li> <li><a href="/author/DELS/">Division on Earth and Life Studies</a></li> <li><a href="/author/DEPS/">Division on Engineering and Physical Sciences</a> </li> <li><a href="/author/GRP/">Gulf Research Program</a></li> <li><a href="/author/HMD/">Health and Medicine Division</a></li> <li><a href="/author/PGA/">Policy and Global Affairs</a></li> <li><a href="/author/TRB/">Transportation Research Board</a></li> <li><hr style="margin-left: 21px; margin-right: 20px;"></li> <li><a href="/author/NAS">National Academy of Sciences</a></li> <li><a href="/author/NAE">National Academy of Engineering</a></li> <li><a href="/author/NAM">National Academy of Medicine</a></li> </ul> <ul class="browse-col" > <li class="dropdown-header"><a href="/topic">Publications by Topic</a></li> <li><a href="/topic/276">Agriculture</a></li> <li><a href="/topic/277">Behavioral and Social Sciences</a></li> <li><a href="/topic/292">Biography and Autobiography</a></li> <li><a href="/topic/278">Biology and Life Sciences</a></li> <li><a href="/topic/279">Computers and Information Technology</a></li> <li><a href="/topic/280">Conflict and Security Issues</a></li> <li><a href="/topic/281">Earth Sciences</a></li> <li><a href="/topic/282">Education</a></li> <li><a href="/topic/283">Energy and Energy Conservation</a></li> <li><a href="/topic/284">Engineering and Technology</a></li> </ul> <ul class="browse-col"> <li><br> </li> <li><a href="/topic/285">Environment and Environmental Studies</a></li> <li><a href="/topic/287">Food and Nutrition</a></li> <li><a href="/topic/288">Health and Medicine</a></li> <li><a href="/topic/289">Industry and Labor</a></li> <li><a href="/topic/290">Math, Chemistry, and Physics</a></li> <li><a href="/topic/423">Policy for Science and Technology</a></li> <li><a href="/topic/293">Space and Aeronautics</a></li> <li><a href="/topic/478">Surveys and Statistics</a></li> <li><a href="/topic/294">Transportation and Infrastructure</a></li> </ul> <ul class="browse-col vertical-divider"> <li class="dropdown-header"><br></li> <li><a href="/collection">Searchable Collections</a></li> <li><a href="/new">New Releases</a></li> </ul> </li> </ul> </li> <li class="dropdown"> <a class="dropdown-toggle" data-toggle="dropdown" href="#"><span class="icon-menu" data-icon=""> </span> MyNAP <span class="caret"></span></a> <ul class="dropdown-menu"> <li><a href="/login.php" class="userOff"><span class="icon-menu" data-icon=""> </span> Login</a></li> <li><a href="/login.php?action=new" class="register"><span class="icon-menu" data-icon=""> </span> Register</a></li> </ul> </li> <li class="naphelp"><a href="/content/help"><span class="icon-menu" data-icon=""> </span> HELP</a></li> <li> <a href="/cart/cart.php"> <span class="icon-menu" data-icon=""> </span> Cart </a> </li> </ul> </div> <div class="global-nav__menus"> <div class="global-nav__menus--upper"> <div class="global-nav__controls"> <div class="controls__social"><!----></div> <button aria-label="Search" class="controls__search" onclick="toggleSearch()"><span class="search__label"> Search </span> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""> <path d="M13.8 21.6a7.8 7.8 0 110-15.6 7.8 7.8 0 010 15.6zm0-2a5.8 5.8 0 100-11.6 5.8 5.8 0 000 11.6z"></path> <path d="M17.899 19.325l1.414-1.414 4.322 4.322-1.414 1.414z"></path> </svg> </button> <button class="controls__menu"> Menu <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""> <path fill-rule="evenodd" d="M8 23h15v-2.5H8V23zm0-6.25h15v-2.5H8v2.5zM8 8v2.5h15V8H8z"></path> </svg> </button> <button class="controls__nav" onclick="toggleMenu(); myMenu(this)"> Global Menu <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="menu-off"> <path fill-rule="evenodd" d="M8 23h15v-2.5H8V23zm0-6.25h15v-2.5H8v2.5zM8 8v2.5h15V8H8z"></path> </svg> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="menu-on"><path fill-rule="evenodd" d="M20.728 8l1.818 1.818-5.455 5.455 5.455 5.455-1.818 1.818-5.455-5.455-5.455 5.455L8 20.728l5.455-5.455L8 9.818 9.818 8l5.455 5.455z"></path></svg> <!----></button> </div> </div> <div class="global-nav__menus--lower" > <nav role="navigation" class="global-nav__links " > <ul role="menu"> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/about" target="_self" role="menuitem" class="global-nav__link"><span> About Us </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"> <ul class="global-nav__links-widget" type="nas-nav-links-widget"> <li><a rel="" title="" href="https://www.nationalacademies.org/about" target="_self"> About Us </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/organization" target="_self"> Our Program Divisions </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/the-national-academies" target="_self"> Our Three Academies </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/leadership" target="_self"> Leadership </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/newsroom" target="_self"> Newsroom </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/ocga" target="_self"> Government Affairs </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/giving" target="_self"> Giving </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/statement-on-diversity-and-inclusion" target="_self"> Statement on Diversity and Inclusion </a></li> </ul> </div> <div class="dropdown__right"> <p> <p>The National Academies of Sciences, Engineering, and Medicine are private, nonprofit institutions that provide expert advice on some of the most pressing challenges facing the nation and world. Our work helps shape sound policies, inform public opinion, and advance the pursuit of science, engineering, and medicine.</p> </p> <a rel="" title="Learn More" href="https://www.nationalacademies.org/about" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> Learn More <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/events" target="_self" role="menuitem" class="global-nav__link"><span> Events </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"><span class="dropdown__left--default"><img src="https://nasemoceprodcomm-nationalacademies.cec.ocp.oraclecloud.com/content/published/api/v1.1/assets/CONTC88C4C0CA7CE4E7FA8630DEF6E1B3A8F/native/NAS+Building.jpg?cb=_cache_a061&channelToken=d2dc71410b56f1d61406108f1738d0f4" alt=""></span></div> <div class="dropdown__right"> <p> <p>Throughout any given year, the National Academies convene hundreds of conferences, workshops, symposia, forums, roundtables, and other gatherings that attract the finest minds in academia and the public and private sectors. These venues for discussion and debate are essential for allowing the scientific process to unfold.</p> </p> <a rel="" title="See All Events" href="https://www.nationalacademies.org/events" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> See All Events <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/our-work" target="_self" role="menuitem" class="global-nav__link"><span> Our Work </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"> <ul class="global-nav__links-widget" type="nas-nav-links-widget"> <li><a rel="" title="" href="https://www.nationalacademies.org/our-work" target="_self"> Our Work </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/our-study-process" target="_self"> Our Study Process </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/conflict-of-interest-policies-and-procedures" target="_self"> Conflict of Interest Policies and Procedures </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/institutional-policies-and-procedures/project-comments-and-information" target="_self"> Project Comments and Information </a></li> </ul> </div> <div class="dropdown__right"> <p> <p>The National Academies of Sciences, Engineering, and Medicine are the nation's pre-eminent source of high-quality, objective advice on science, engineering, and health matters. <br> <br> Top experts participate in our projects, activities, and studies to examine and assemble evidence-based findings to address some of society's greatest challenges.</p> </p> <a rel="" title="Learn More" href="https://www.nationalacademies.org/our-work" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> Learn More <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/publications" target="_self" role="menuitem" class="global-nav__link"><span> Publications </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"> <ul class="global-nav__links-widget" loadchildren="true" type="nas-nav-links-widget"> <li><a rel="" title="" href="https://nap.nationalacademies.org/" target="_self"> Read Our Expert Reports and Published Proceedings </a></li> <li><a rel="" title="" href="https://www.pnas.org/" target="_self"> Explore PNAS, the Flagship Scientific Journal of NAS </a></li> <li><a rel="" title="" href="http://pubsindex.trb.org/" target="_self"> Access Transportation Research Board Publications </a></li> </ul> </div> <div class="dropdown__right"> <p>Our peer-reviewed reports present the evidence-based consensus of committees of experts. Published proceedings record the presentations and discussions that take place at hundreds of conferences, workshops, symposia, forums, roundtables, and other gatherings every year. And, our prestigious journals publish the latest scientific findings on a wide range of topics.</p> <a rel="" title="Learn More" href="https://www.nationalacademies.org/publications" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> Learn More <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/topics" target="_self" role="menuitem" class="global-nav__link"><span> Topics </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"> <ul class="global-nav__links-widget" type="nas-nav-links-widget"> <li><a rel="" title="" href="https://www.nationalacademies.org/topics/covid-19-resources" target="_self"> Coronavirus Disease 2019 (COVID-19) </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/topics/resources-on-diversity-equity-and-inclusion" target="_self"> Diversity, Equity, and Inclusion </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/topics/climate" target="_self"> Climate </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/topics/economic-recovery" target="_self"> Economic Recovery </a></li> </ul> </div> <div class="dropdown__right"> <p> <p>Discover what the National Academies are doing in various topic areas to strengthen the fields of science, engineering, and medicine and their capacity to contribute to the well-being of our nation and the world.</p> </p> <a rel="" title="See All Topics" href="https://www.nationalacademies.org/topics" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> See All Topics <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/engagement" target="_self" role="menuitem" class="global-nav__link"><span> Engagement </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"><span class="dropdown__left--default"><img src="https://nasemoceprodcomm-nationalacademies.ocecdn.oraclecloud.com/content/published/api/v1.1/assets/CONTC267CED7812243AB86410D548C73D067/native/engagement-nav.jpg?cb=_cache_a061&channelToken=d2dc71410b56f1d61406108f1738d0f4" alt=""></span></div> <div class="dropdown__right"> <p> <p>A powerful way to foster appreciation for the impact of science and critical and innovative thinking is through art and the humanities. Learn more about the National Academies’ various programs designed to connect, engage, and inspire.</p> </p> <a rel="" title="Learn More" href="https://www.nationalacademies.org/engagement" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> Learn More <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> <li role="none" class="has-dropdown"><a href="https://www.nationalacademies.org/opportunities" target="_self" role="menuitem" class="global-nav__link"><span> Opportunities </span></a> <div class="global-nav__dropdown"> <div class="dropdown__left"> <ul class="global-nav__links-widget" type="nas-nav-links-widget"> <li><a rel="" title="" href="https://www.nationalacademies.org/careers" target="_self"> Careers </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/fellowships-and-grants" target="_self"> Fellowships and Grants </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/volunteers" target="_self"> Volunteers </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/sponsors" target="_self"> Sponsors </a></li> <li><a rel="" title="" href="https://www.nationalacademies.org/about/members" target="_self"> Members </a></li> </ul> </div> <div class="dropdown__right"> <p>Make a real impact on the scientific, engineering, and health-related challenges facing society. Whether as a sponsor or donor, a member or volunteer, or an employee or fellow, you can make a difference.</p> <a rel="" title="Learn More" href="https://www.nationalacademies.org/opportunities" target="_self" role="menuitem" class="button button--secondary button--outline button--icon"> Learn More <span class="button__icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class="w2 h2"> <path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path> <path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path> </svg> </span></a></div> </div> </li> </ul> </nav> </div> </div> </div> <!----></header> <a name="content"></a> <script> $(".controls__menu").click(function(){ $("#mobile-menu").toggle(); }); function toggleMenu() { var element = document.getElementById("lowerMenu"); element.classList.toggle("menu-is-hidden"); } function myMenu(x) { x.classList.toggle("invert"); } function toggleSearch() { var element = document.getElementById("lowerMenu"); element.classList.toggle("search-is-hidden"); } </script> <!--accordion--> <script> var acc = document.getElementsByClassName("accordion"); var i; for (i = 0; i < acc.length; i++) { acc[i].addEventListener("click", function() { this.classList.toggle("active"); var panel = this.nextElementSibling; if (panel.style.maxHeight) { panel.style.maxHeight = null; } else { panel.style.maxHeight = panel.scrollHeight + "px"; } }); } </script> <div class="container"> <script language="JavaScript" type="text/javascript"> <!-- // set up formatting variables var phrase_in_context='<font size=1>Key Phrase In Context: '; var p_i_c_table='<table width="97%" bgcolor=darkgray cellpadding=5 cellspacing=1 id=skimresults><tbody>'; var newwindow; function goe(url) { newwindow=window.open(url,name,'height=400,scrollbar=yes,left=100,top=70,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {newwindow.focus()} } var pewwindow; function poe(url) { pewwindow=window.open(url,name,'width=700;scrollbar=yes,left=80,top=90,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {pewwindow.focus()} } var yewwindow; function yoe(url) { yewwindow=window.open(url,name,'height=400,scrollbar=yes,left=120,top=110,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {yewwindow.focus()} } var mewwindow; function moe(url) { mewwindow=window.open(url,name,'height=400,scrollbar=yes,left=140,top=130,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {mewwindow.focus()} } var msewwindow; function msoe(url) { msewwindow=window.open(url,name,'height=400,scrollbar=yes,left=90,top=110,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {msewwindow.focus()} } var sewwindow; function soe(url) { sewwindow=window.open(url,name,'height=400,scrollbar=yes,left=160,top=150,resizable=yes,scrollbars=yes,toolbar=yes,status=yes,menubar=yes,location=yes'); if (window.focus) {sewwindow.focus()} } function show_engines() { document.getElementById("infospace").innerHTML = '<table cellpadding=0 cellspacing=4><tbody><tr><td colspan=2 align=center><div class="optitle">WebResearch Builder ON</div>' + '<font size=1>[<a href="javascript:click_phrases();">Phrase Click</a>] [<a href="javascript:show_start();">Skim View</a>]<br> ' + 'Click on term to display Web Searches<hr size=1 color=darkblue></td></tr>' + '<tr><td valign=top style=\"linkphrase\"><ul></ul></td><td valign=top style=\"linkphrase\"><ul></ul></td></tr>' + '</tbody></table>'; document.getElementById("text2").innerHTML = '<font size=2><br><b>Special WebResearch Feature</b><br><br>Click on the key terms to the left (extracted from this report) to display lists of two-phrase, ' + 'targetted searches to send to Google for Web-wide searches, or to the National Academies Press Discovery Engine for searching more than 3300 publications from the National Academies.<br>' + '<br>This allows one-click searches on combination terms like <br><br><b>"" and ""</b></font>' + '<br>NOTE: Clicking on Google or National Academies Press will open a new window displaying search results.<br><br></font>' + '<hr size=1 color=darkblue>Search for <font size=2><b></b></font> ...</b><hr size=1 color=darkblue>' + '<table bgcolor=white width=100%><tbody>' + '</tbody></table><hr color=darkblue size=1>'; } function click_phrases() { document.getElementById("infospace").innerHTML = '<table cellpadding=0 cellspacing=4><tbody><tr><td colspan=2 align=center><div class=\"optitle\">Phrase Click ON</div><font size=1>[<a href=\"#top\" onclick=\"rollover_phrases()\">Phrase Rollover' + '</a>] [<a href=\"javascript:DoDisplay(\'adversarial_attacks\')\;\">WebResearch Builder</a>] [<a href=\"javascript:show_start()\;\">Skim View</a>]<br>&bull\; key term (#pages)' + '</font><hr size=1 color=darkblue></td></tr><tr><td valign=top><ul></ul></td><td valign=top><ul></ul></td></tr></tbody></table>'; } function show_start() { // should return to the "chapter skim"; // html-preps top N number to have in infospace var displayit=''; for (x=0;x<TopPhrases.length;x++){ // for (x=0;x<30;x++){ if (x==5|x==10|x==15|x==20|x==25|x==30|x==35|x==40|x==45|x==50|x==55|x==60|x==65|x==70){ displayit=displayit + '<div class=ider><br></div>'; } var thisphrase=TopPhrases[x]; var passphrase=thisphrase.replace('_','+'); var showphrase=thisphrase.replace('_',' '); displayit= displayit + '<li class="xy"><a href="javascript:x_' + thisphrase + '();">' + showphrase + '</a></li>'; } document.getElementById("text2").innerHTML = '<table width=\"97%\" bgcolor=darkgray cellpadding=5 cellspacing=1 id=\"skimresults\"><tbody><tr bgcolor=white><td valign=top><div class=fromblock>From ' + 'page 102...</div><div class=foundblock>... <font class=\"btext\">The employment of AI-enabled systems can have significant benefits in augmenting the capabilities ' + 'of the warfighter, but there are other risks inherent in the use of AI-enabled systems that must be considered. In particular, this chapter answers the ' + 'second of the committee\'s three primary tasks, \"consider examples of AI corruption under operational conditions and against malicious cyberattacks.\" ' + '5.1 INTRODUCTION An AI-enabled system includes technical risks that must be considered during the test and evaluation (T&E)</font></div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=102\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... <font class=\"btext\">In addition, OT&E performance tests ' + 'can ferret out operational environmental degrading ef fects, which may motivate model retraining using new datasets. Some attacks, referred to as backdoor ' + 'attacks, only manifest once the adversary triggers them.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... <font ' + 'class=\"btext\">Instead, AI models learn to make predictions based on pat terns and relationships specific to the training data. As a result, they do ' + 'not generalize well to novel data, and even a bit of extra signal noise in a sensor collection, which might be invisible to the human eye, can potentially ' + 'confuse an AI system.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... <font ' + 'class=\"btext\">Thus, while this report focuses specifically on the question of adversarial attacks against the AI components, it is important to con tinue ' + 'to test and evaluate AI-enabled systems against a broad spectrum of adversaries and attacks in a realistic operational context. In addition to development ' + 'and test datasets, the collection and use of operational data, especially in training AI compo nents, is key to mitigating threats to the AI component ' + 'and the system as a whole.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... <font ' + 'class=\"btext\">This may be in the form of statistical measures such as precision and recall, reduction of a performance envelope required for a mission ' + 'objective, or in a violation of the system\'s security requirements, such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against ' + 'AI Systemsa Adversarial examples: These are inputs to a machine learning model that are specifically designed to mislead the model and cause it to make ' + 'incorrect predictions. Adversarial examples can be created by adding small, imperceptible perturbations to the input data designed to fool the model.</font></div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... <font class=\"btext\">Thus, all traditional testing for ' + 'adversarial attacks against the attack surface that has been previously defined is still required. Defensive mechanisms that obvi ate or limit the capability ' + 'of an adversary to take advantage of an attack surface are also the first line of defense for AI-enabled components.</font></div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... <font class=\"btext\">AI model inversion attacks refer ' + 'to techniques that aim to reverse engineer the internal workings of an AI model. These attacks are a type of adversarial attack in which an attacker seeks ' + 'to reconstruct the input data or features used to train a model or to generate synthetic inputs that will produce a desired output from the model.</font></div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... <font class=\"btext\">In summary, the attack surface ' + 'of AI-enabled systems encompasses all tra ditional attack surfaces inherent in software-intensive systems -- especially those of ML-enabled systems -- ' + 'but has the additional considerations of AI corruption through the life cycle of the AI system, including the data used in training, test, and operations, ' + 'and the access to the details in any configuration or model that is part of the AI component. 5.5 RISK OF ADVERSARIAL ATTACKS It can be useful to divide ' + 'the risk of adversarial attacks into the different levels of integration of the AI-enabled system.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... <font ' + 'class=\"btext\">Note, however, that over-training in the adversarial space could create vulnerabilities in the inversion attack of the resulting model, ' + 'thus exposing which attacks and training sets are used to an adversary with limited access to the resulting deployed system. Recognizing the planned mission ' + 'objectives is also important when testing these components.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 111...</div><div class=foundblock>... <font ' + 'class=\"btext\">Palat, and A Sinha, 2022, \"Measuring AI Systems Beyond Accuracy,\" paper presented at AAAI Spring Symposium Series Workshop on AI Engineering: ' + 'Creating Scalable, Human-Centered and Robust AI Systems, https://doi.org/10.48550/ arXiv.2204.04211.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=111\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 112...</div><div class=foundblock>... <font ' + 'class=\"btext\">In addition, network security, data protection, encryp tion, and other classic defenses must be integrated at this point into the operational ' + 'system and tested as part of the OT&E process. Finally, the risk of adversarial attack during deployment is increased over the traditional software-intensive ' + 'systems by creating an additional attack surface FIGURE 5-2 The operational environment for ML-enabled systems.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=112\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... <font ' + 'class=\"btext\">The execution of Zero Trust provides targeteers trusted, tagged, and labeled data so they can confi dently employ and share it with trusted ' + 'partners, assured that the data is protected, secure, and accessed by only the people who need it when they need it, using least privilege principles.5 ' + ' 5 DoD, 2022, \"Zero Trust Strategy.\" </font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... <font ' + 'class=\"btext\">In addition, using specific APIs to control data flow and access to the AI life cycle and deployed environments will limit the ability ' + 'of the adversary to propagate attacks to multiple operational systems. Note this is also a trade-off to a dynamic approach to DevSecOps that would enable ' + 'near real-time updates to operational models based on retraining and updates on changing deployed environments as well as the use of operational data in ' + 'the development and training of new models.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... <font ' + 'class=\"btext\">Finding 5-3: The impact of adversarial attacks on human-AI enabled systems has not been well understood. At present, the DoD Zero Trust ' + 'Strategy is only being implemented on en terprise systems.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 116...</div><div class=foundblock>... <font ' + 'class=\"btext\">The Department\'s ability to protect, and by extension, DoD personnel against the array of increasingly sophisticated cybersecurity threats ' + 'depends on it. 5.7 ROBUST AND SECURE AI MODELS A common approach for increasing the robustness and security of AI models is the incorporation of monitoring ' + 'or watchdog systems that compare the output of an AI-enabled system to pre-defined operational limits.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=116\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... <font ' + 'class=\"btext\">In particular, attacks that require unimpeded access to an underlying AI model are unlikely to be practical with traditional protections ' + 'and mitigations inherent in deployed DAF systems. Nevertheless, as stated 7 Variations regarding adversarial attacks were discussed in Section 3; in ' + 'this section robustness also includes both adversarial and non-adversarial variations in the environment.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... <font ' + 'class=\"btext\">These may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without ' + 'direct access. Even attacks that require unimpeded access to the AI model may be successful due to the transferability of adversarial attacks.</font></div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 119...</div><div class=foundblock>... <font class=\"btext\">In many cases, sensitive data would ' + 'need to be sanitized to prevent training data from exposure during model inver sion or when testing new types of AI-enabled systems. However, current ' + 'approaches to sanitization do not effectively support the trade-off between the effective training of the system and the possibility of leaking sensitive ' + 'data.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=119\');\">Read the entire ' + 'page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 120...</div><div class=foundblock>... <font class=\"btext\">120 ' + ' Test and E va l u at i o n C h a l l e n g e s in AI - E n a b l e d S y s t e m s Conclusion 5-1: Promising areas of research that will improve ' + 'the mitigation of adversarial AI include techniques for data sanitization, quantifiable uncertainty, and certifiable robustness. Additionally, Chapter ' + '6 discusses additional emerging AI technologies and promising areas of research.</font></div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=120\');\">Read ' + 'the entire page → </a></div></td></tr>'; document.getElementById("texttitle").innerHTML = ''; document.getElementById("linkphrase").innerHTML = displayit; } var AllPhrases=new Array(); AllPhrases.push('adversarial_attacks'); AllPhrases.push('zero_trust'); AllPhrases.push('training_data'); AllPhrases.push('operational_conditions'); AllPhrases.push('operational_data'); AllPhrases.push('model_inversion'); AllPhrases.push('attacks_involve'); AllPhrases.push('machine_learning'); AllPhrases.push('trust_strategy'); AllPhrases.push('supply_chain'); AllPhrases.push('adversarial_attack'); AllPhrases.push('traditional_attack'); AllPhrases.push('operational_environments'); AllPhrases.push('software_component'); AllPhrases.push('learning_model'); AllPhrases.push('inversion_attacks'); AllPhrases.push('input_data'); AllPhrases.push('dod_zero'); AllPhrases.push('cyberattack_surface'); AllPhrases.push('require_unimpeded'); AllPhrases.push('adversarial_examples'); AllPhrases.push('adversarial_manipulation'); AllPhrases.push('data_sanitization'); AllPhrases.push('distribution_shifts'); AllPhrases.push('test_datasets'); AllPhrases.push('network_security'); AllPhrases.push('unimpeded_access'); AllPhrases.push('model_performance'); AllPhrases.push('attack_surfaces'); AllPhrases.push('deployed_environments'); // establish top phrases var TopPhrases=new Array(); for (x=0;x<30;x++){ TopPhrases.push(AllPhrases[x]); } function x_adversarial_attacks() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"adversarial attacks"</b> shown in context from 9 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_adversarial_attacks" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... <b>adversarial attacks</b>. However, the Department ' + 'of the Air Force (DAF)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... The DAF ' + 'will have to depend on cy ber defenses to broadly restrict <b>adversarial attacks</b> by denying access to data (training and testing) and queries (to ' + 'protect against inversion and privacy attacks)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... \"[e] ' + 'valuates whether autonomous and semi-autonomous weapon systems under DOT&E oversight have met standards for rigorous V&V and T&E in realistic operational ' + 'conditions, including potential adversary action, to provide sufficient confidence that the probability and consequences of failures have been minimized.\"2 ' + 'This latest policy on autonomous weapon systems illustrates the nature of placing <b>adversarial attacks</b> against AI-enabled systems in the context ' + 'of all threats in realistic operating conditions.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Thus, ' + 'while this report focuses specifically on the question of <b>adversarial attacks</b> against the AI components, it is important to con tinue to test and ' + 'evaluate AI-enabled systems against a broad spectrum of adversaries and attacks in a realistic operational context. In addition to development and test ' + 'datasets, the collection and use of operational data, especially in training AI compo nents, is key to mitigating threats to the AI component and the system ' + 'as a whole.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read the entire ' + 'page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 105</i></div></td></tr><tr bgcolor=white><td valign=top><div ' + 'class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such as precision and recall, reduction ' + 'of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, such as maintaining the secrecy BOX ' + '5-1 Some Examples of <b>Adversarial Attacks</b> Against AI Systemsa Adversarial examples: These are inputs to a machine learning model that are specifically ' + 'designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created by adding small, imperceptible perturbations ' + 'to the input data designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... Thus, ' + 'all traditional testing for <b>adversarial attacks</b> against the attack surface that has been previously defined is still required. Defensive mechanisms ' + 'that obvi ate or limit the capability of an adversary to take advantage of an attack surface are also the first line of defense for AI-enabled components.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... In summary, the attack surface of AI-enabled systems ' + 'encompasses all tra ditional attack surfaces inherent in software-intensive systems -- especially those of ML-enabled systems -- but has the additional ' + 'considerations of AI corruption through the life cycle of the AI system, including the data used in training, test, and operations, and the access to the ' + 'details in any configuration or model that is part of the AI component. 5.5 RISK OF <b>ADVERSARIAL ATTACKS</b> It can be useful to divide the risk of <b>adversarial ' + 'attacks</b> into the different levels of integration of the AI-enabled system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... These ' + 'attacks are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Finding 5-2: Ongoing research on <b>adversarial ' + 'attacks</b> on AI-enabled systems focus on performance on benchmark datasets which are inadequate for simulating operational attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... It appears that as robustness to <b>adversarial attacks</b> ' + 'is improved, the performance often goes down. Even on benchmark datasets, the trade-off between potential performance reduction and im proved robustness ' + 'is not understood.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 115</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... In particular, attacks that require unimpeded access to an underlying AI ' + 'model are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Nevertheless, as stated 7 Variations ' + 'regarding <b>adversarial attacks</b> were discussed in Section 3; in this section robustness also includes both adversarial and non-adversarial variations ' + 'in the environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. Even ' + 'attacks that require unimpeded access to the AI model may be successful due to the transferability of <b>adversarial attacks</b>.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... 2. The scenario where the adversary attempts to compromise ' + 'our cyber secu rity and then uses a variety of <b>adversarial attacks</b>.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_zero_trust() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"zero trust"</b> shown in context from 4 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_zero_trust" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Schneider, and N Carlini, 2022, \"DoD <b>Zero Trust</b> ' + 'Strat egy,\" Washington, DC: Department of Defense, https://dodcio.defense.gov/Portals/0/Documents/ Library/DoD-ZTStrategy.pdf.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... 5.6 NETWORK SECURITY AND <b>ZERO TRUST</b> IMPLICATIONS ' + 'Modern network security within DoD relies on the principles of <b>zero trust</b> (ZT)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... <b>Zero ' + 'trust</b> is the term for an \"evolving set of cybersecurity paradigms that move defenses from static, network-based perimeters to focus on users, as sets, ' + 'and resources.\" <b>Zero trust</b> uses continuous multi-factor authentication, micro-segmentation, advanced encryption, endpoint security, analytics, ' + 'and robust auditing, among other capabilities, to fortify data, applications, assets, and services to deliver cyber resiliency. DoD is evolving to become ' + 'a more agile, more mobile, cloud-supported workforce, collaborating with the entirety of the DoD enterprise, including federal and non-federal organizations ' + 'and mission partners working on various missions.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 3 more on page 113</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... This is primarily because the nature of supply chain attacks on the training, ' + 'test, and validation data for AI models require the same level of authentication and autho rization as access to the algorithms and models themselves. ' + 'This is a fundamental principle of <b>zero trust</b> that all elements of the system, data, network, software, and interfaces be authenticated and authorized ' + 'by the role of the user or software with access to these capabilities.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... Finding ' + '5-3: The impact of adversarial attacks on human-AI enabled systems has not been well understood. At present, the DoD <b>Zero Trust</b> Strategy is only ' + 'being implemented on en terprise systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... should ' + 'ap ply the DoD <b>Zero Trust</b> Strategy to all DAF artificial intelligence–enabled systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_training_data() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"training data"</b> shown in context from 6 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_training_data" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... In particular, physically-based attacks using camouflage, ' + 'concealment, or deception are possible avenues to biasing, denying, and poisoning <b>training data</b>, training labels, and operational data. Particularly ' + 'insidious are back door attacks, which poison the <b>training data</b> (without changing the training labels)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... 5.2 GENERAL ' + 'RISKS OF AI-ENABLED SYSTEMS From a T&E perspective, the areas of an AI-enabled system that must be pro tected are the availability, integrity, and confidentiality ' + 'and privacy of the AI system: • Availability and integrity of the AI output • The integrity of the AI model, data, and software • Confidentiality ' + 'and privacy of the model and <b>training data</b> Several risks are inherent to using AI in operational environments. Awareness is the first step in mitigating ' + 'such risks, so the committee discusses here some main challenges that have been identified.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... AI is ' + 'dependent on its <b>training data</b>, and its predictions are only as good as the data it has been trained on. Limited data in some scenarios can lead ' + 'to inherent biases and risks.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 104</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Poisoning attacks: These attacks involve introducing malicious or misleading ' + 'data into a ma chine learning model\'s <b>training data</b> in an attempt to corrupt the model\'s output. Evasion attacks: These attacks involve manipulating ' + 'the input data specifically to evade detec tion by a machine learning model that is being used for security or fraud detection purposes.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... Yet another attack that is usually not part of a traditional ' + 'attack surface is the manipulation of <b>training data</b> to an ML component to cause the model to learn a response beneficial to the adversary in an ' + 'operational deployment. This can be accomplished not only in the initial training of the model but during retraining when the model is updated to respond ' + 'to updated environmental conditional post-deployment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... Model ' + 'inversion attacks are a potential concern because they could allow an attacker to learn sensitive information about the <b>training data</b> or the training ' + 'process itself, which could exploit the model\'s vulnerabilities or craft adversarial examples that can fool the model. Model inversion attacks can be ' + 'particularly dangerous when applied to models used in high-stakes situations, such as DoD </div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... Note that ' + 'this approach may increase the risk of adversary manipulation of the re<b>training data</b> to drift the model to be more beneficial to the adver sary ' + 'once the retrained model is deployed. This is an inherent trade-off in retraining with operational data that includes adversarial input.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 119...</div><div class=foundblock>... In many cases, sensitive data would need to be sanitized ' + 'to prevent <b>training data</b> from exposure during model inver sion or when testing new types of AI-enabled systems. However, current approaches to ' + 'sanitization do not effectively support the trade-off between the effective training of the system and the possibility of leaking sensitive data.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=119\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_operational_conditions() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"operational conditions"</b> shown in context from 5 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_operational_conditions" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 102...</div><div class=foundblock>... The employment of AI-enabled systems can have significant ' + 'benefits in augmenting the capabilities of the warfighter, but there are other risks inherent in the use of AI-enabled systems that must be considered. ' + 'In particular, this chapter answers the second of the committee\'s three primary tasks, \"consider examples of AI corruption under <b>operational conditions</b> ' + 'and against malicious cyberattacks.\" 5.1 INTRODUCTION An AI-enabled system includes technical risks that must be considered during the test and evaluation ' + '(T&E)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=102\');\">Read the entire page ' + '→ </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... For instance, having no ' + 'training data from below-freezing weather conditions means the AI model will not be accurate in such <b>operational conditions</b>. Additionally, AI implementations ' + 'are very sensitive to distribution shifts, which often happen gradually and slowly degrade the performance of an AI element, in dependent system, or joint ' + 'cognitive system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... \"[e] ' + 'valuates whether autonomous and semi-autonomous weapon systems under DOT&E oversight have met standards for rigorous V&V and T&E in realistic <b>operational ' + 'conditions</b>, including potential adversary action, to provide sufficient confidence that the probability and consequences of failures have been minimized.\"2 ' + 'This latest policy on autonomous weapon systems illustrates the nature of placing adversarial attacks against AI-enabled systems in the context of all ' + 'threats in realistic operating conditions.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... Thus, ' + 'removing traditional vulnerabilities and adding robust protections to systems containing AI components can limit an adversary\'s ability to influence, ' + 'deny, degrade, or corrupt the functions of the AI component. In all cases, the limitation of access by the adversary to the component is considered the ' + 'first line of defense in preventing AI corruption in <b>operational conditions</b>.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... Recommendation ' + '5-1: The Department of the Air Force (DAF) should fund research activities that investigate the trade-offs between model resilience to adversarial attack ' + 'and model performance under <b>operational conditions</b>.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_operational_data() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"operational data"</b> shown in context from 4 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_operational_data" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 102...</div><div class=foundblock>... of any system that incorporates AI elements. The ' + 'most likely risks to AI-enabled systems are enabled through cyber access to the AI components, be they the training or <b>operational data</b>, the model, ' + 'the software implementing the component, or the output of the AI-enabled system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=102\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... In particular, ' + 'physically-based attacks using camouflage, concealment, or deception are possible avenues to biasing, denying, and poisoning training data, training labels, ' + 'and <b>operational data</b>. Particularly insidious are back door attacks, which poison the training data (without changing the training labels)</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Thus, while this report focuses specifically on the question ' + 'of adversarial attacks against the AI components, it is important to con tinue to test and evaluate AI-enabled systems against a broad spectrum of adversaries ' + 'and attacks in a realistic operational context. In addition to development and test datasets, the collection and use of <b>operational data</b>, especially ' + 'in training AI compo nents, is key to mitigating threats to the AI component and the system as a whole.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... In addition, ' + 'using specific APIs to control data flow and access to the AI life cycle and deployed environments will limit the ability of the adversary to propagate ' + 'attacks to multiple operational systems. Note this is also a trade-off to a dynamic approach to DevSecOps that would enable near real-time updates to operational ' + 'models based on retraining and updates on changing deployed environments as well as the use of <b>operational data</b> in the development and training ' + 'of new models.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read the ' + 'entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... Note that ' + 'this approach may increase the risk of adversary manipulation of the retraining data to drift the model to be more beneficial to the adver sary once ' + 'the retrained model is deployed. This is an inherent trade-off in retraining with <b>operational data</b> that includes adversarial input.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_model_inversion() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"model inversion"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_model_inversion" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... <b>Model inversion</b> attacks: These attacks involve ' + 'attempting to recover sensitive information about the data used to train a machine learning model by manipulating the model\'s output and using it to infer ' + 'information about the input data. Also known as a data inference attack.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... AI <b>model ' + 'inversion</b> attacks refer to techniques that aim to reverse engineer the internal workings of an AI model. These attacks are a type of adversarial attack ' + 'in which an attacker seeks to reconstruct the input data or features used to train a model or to generate synthetic inputs that will produce a desired ' + 'output from the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... <b>Model ' + 'inversion</b> attacks are a potential concern because they could allow an attacker to learn sensitive information about the training data or the training ' + 'process itself, which could exploit the model\'s vulnerabilities or craft adversarial examples that can fool the model. <b>Model inversion</b> attacks ' + 'can be particularly dangerous when applied to models used in high-stakes situations, such as DoD </div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include <b>model inversion</b> attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. ' + 'Even attacks that require unimpeded access to the AI model may be successful due to the transferability of adversarial attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_attacks_involve() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"attacks involve"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_attacks_involve" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Poisoning attacks: These <b>attacks involve</b> introducing ' + 'malicious or misleading data into a ma chine learning model\'s training data in an attempt to corrupt the model\'s output. Evasion attacks: These <b>attacks ' + 'involve</b> manipulating the input data specifically to evade detec tion by a machine learning model that is being used for security or fraud detection ' + 'purposes.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read the entire ' + 'page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Model inversion attacks: ' + 'These <b>attacks involve</b> attempting to recover sensitive information about the data used to train a machine learning model by manipulating the model\'s ' + 'output and using it to infer information about the input data. Also known as a data inference attack.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 106</i></div></td></tr>' + '</tbody></table>' + ''; } function x_machine_learning() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"machine learning"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_machine_learning" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such ' + 'as precision and recall, reduction of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, ' + 'such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against AI Systemsa Adversarial examples: These are inputs to a <b>machine ' + 'learning</b> model that are specifically designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created ' + 'by adding small, imperceptible perturbations to the input data designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Poisoning ' + 'attacks: These attacks involve introducing malicious or misleading data into a ma chine learning model\'s training data in an attempt to corrupt the ' + 'model\'s output. Evasion attacks: These attacks involve manipulating the input data specifically to evade detec tion by a <b>machine learning</b> model ' + 'that is being used for security or fraud detection purposes.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 3 more on page 106</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... Often the theoretical maximum performance of an ML com ponent is tuned ' + 'to some imagined optimal point, but operational requirements may dictate a different point to tune the trade-offs inherent in <b>machine learning</b>. ' + 'For example, it may be necessary to have a high-precision result at the expense of recall for a fix and target application, but for a surveillance mission, ' + 'a high recall result may better fit operational needs.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_trust_strategy() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"trust strategy"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_trust_strategy" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... It will also ensure that any adversary damage is ' + 'quickly contained and remediated if a device, network, user, or credential is compromised. As the Zero <b>Trust Strategy</b> states: Trusted Interoperability ' + 'Data for Warfighters: Military targeteers need secure access to data at the speed of relevance they can use and trust.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... The execution ' + 'of Zero Trust provides targeteers trusted, tagged, and labeled data so they can confi dently employ and share it with trusted partners, assured that ' + 'the data is protected, secure, and accessed by only the people who need it when they need it, using least privilege principles.5 5 DoD, 2022, \"Zero ' + '<b>Trust Strategy</b>.\" </div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... Finding ' + '5-3: The impact of adversarial attacks on human-AI enabled systems has not been well understood. At present, the DoD Zero <b>Trust Strategy</b> is only ' + 'being implemented on en terprise systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... should ' + 'ap ply the DoD Zero <b>Trust Strategy</b> to all DAF artificial intelligence–enabled systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_supply_chain() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"supply chain"</b> shown in context from 5 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_supply_chain" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... Beyond the traditional cyber risks that would enable ' + 'the adversary to access the AI component, other risks are unique to the operation of the AI component. These involve the data <b>supply chain</b> (training ' + 'and test data)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read the ' + 'entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... The consideration ' + 'of adversarial AI should be in addition to all traditional cyber threats associated with more traditional integrated systems. Software vulner abilities, ' + '<b>supply chain</b> vulnerabilities, insider threats, network vulnerabilities, denial of service attacks, privilege escalation, and root of trust attacks ' + 'are just a few of the traditional threats that remain in AI-enabled systems and must be addressed.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... In addition ' + 'to the cyberattack surface, the AI component may enable an at tack surface beyond the traditional cyberattack surface. This access is due to the dependencies ' + 'on data within the deployed environment and in the backend infra structure, the <b>supply chain</b> of any AI model in the component, and the potential ' + 'for retraining and adaptation with an adversary-controlled environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... The AI ' + 'component is tested as a standalone system at the most basic level. At this level, the primary risk of adversarial attack is in the <b>supply chain</b> ' + 'of the software and data used to construct the AI component.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... This is ' + 'primarily because the nature of <b>supply chain</b> attacks on the training, test, and validation data for AI models require the same level of authentication ' + 'and autho rization as access to the algorithms and models themselves. This is a fundamental principle of zero trust that all elements of the system, data, ' + 'network, software, and interfaces be authenticated and authorized by the role of the user or software with access to these capabilities.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_adversarial_attack() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"adversarial attack"</b> shown in context from 11 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_adversarial_attack" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... <b>adversarial attack</b>s. However, the Department ' + 'of the Air Force (DAF)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... The DAF ' + 'will have to depend on cy ber defenses to broadly restrict <b>adversarial attack</b>s by denying access to data (training and testing) and queries (to ' + 'protect against inversion and privacy attacks)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... \"[e] ' + 'valuates whether autonomous and semi-autonomous weapon systems under DOT&E oversight have met standards for rigorous V&V and T&E in realistic operational ' + 'conditions, including potential adversary action, to provide sufficient confidence that the probability and consequences of failures have been minimized.\"2 ' + 'This latest policy on autonomous weapon systems illustrates the nature of placing <b>adversarial attack</b>s against AI-enabled systems in the context ' + 'of all threats in realistic operating conditions.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Thus, ' + 'while this report focuses specifically on the question of <b>adversarial attack</b>s against the AI components, it is important to con tinue to test and ' + 'evaluate AI-enabled systems against a broad spectrum of adversaries and attacks in a realistic operational context. In addition to development and test ' + 'datasets, the collection and use of operational data, especially in training AI compo nents, is key to mitigating threats to the AI component and the system ' + 'as a whole.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read the entire ' + 'page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 105</i></div></td></tr><tr bgcolor=white><td valign=top><div ' + 'class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such as precision and recall, reduction ' + 'of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, such as maintaining the secrecy BOX ' + '5-1 Some Examples of <b>Adversarial Attack</b>s Against AI Systemsa Adversarial examples: These are inputs to a machine learning model that are specifically ' + 'designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created by adding small, imperceptible perturbations ' + 'to the input data designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... Thus, ' + 'all traditional testing for <b>adversarial attack</b>s against the attack surface that has been previously defined is still required. Defensive mechanisms ' + 'that obvi ate or limit the capability of an adversary to take advantage of an attack surface are also the first line of defense for AI-enabled components.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... AI model inversion attacks refer to techniques that aim ' + 'to reverse engineer the internal workings of an AI model. These attacks are a type of <b>adversarial attack</b> in which an attacker seeks to reconstruct ' + 'the input data or features used to train a model or to generate synthetic inputs that will produce a desired output from the model.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... In summary, the attack surface of AI-enabled systems ' + 'encompasses all tra ditional attack surfaces inherent in software-intensive systems -- especially those of ML-enabled systems -- but has the additional ' + 'considerations of AI corruption through the life cycle of the AI system, including the data used in training, test, and operations, and the access to the ' + 'details in any configuration or model that is part of the AI component. 5.5 RISK OF <b>ADVERSARIAL ATTACK</b>S It can be useful to divide the risk of <b>adversarial ' + 'attack</b>s into the different levels of integration of the AI-enabled system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... The AI ' + 'component is tested as a standalone system at the most basic level. At this level, the primary risk of <b>adversarial attack</b> is in the supply chain ' + 'of the software and data used to construct the AI component.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 112...</div><div class=foundblock>... In addition, ' + 'network security, data protection, encryp tion, and other classic defenses must be integrated at this point into the operational system and tested as ' + 'part of the OT&E process. Finally, the risk of <b>adversarial attack</b> during deployment is increased over the traditional software-intensive systems ' + 'by creating an additional attack surface FIGURE 5-2 The operational environment for ML-enabled systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=112\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... These ' + 'attacks are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Finding 5-2: Ongoing research on <b>adversarial ' + 'attack</b>s on AI-enabled systems focus on performance on benchmark datasets which are inadequate for simulating operational attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... It appears that as robustness to <b>adversarial attack</b>s ' + 'is improved, the performance often goes down. Even on benchmark datasets, the trade-off between potential performance reduction and im proved robustness ' + 'is not understood.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 3 more on page 115</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... In particular, attacks that require unimpeded access to an underlying AI ' + 'model are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Nevertheless, as stated 7 Variations ' + 'regarding <b>adversarial attack</b>s were discussed in Section 3; in this section robustness also includes both adversarial and non-adversarial variations ' + 'in the environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. Even ' + 'attacks that require unimpeded access to the AI model may be successful due to the transferability of <b>adversarial attack</b>s.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... 2. The scenario where the adversary attempts to compromise ' + 'our cyber secu rity and then uses a variety of <b>adversarial attack</b>s.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_traditional_attack() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"traditional attack"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_traditional_attack" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... These attacks, as described below, expand the <b>traditional ' + 'attack</b> surface and should be considered for the T&E of any AI-enabled system. Vulnerabilities in AI-enabled components may be addressed by limiting ' + 'ad versarial access to this component through traditional separation and protec tion mechanisms.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... So, the ' + 'attack surface of a system that includes AI-enabled compo nents also includes the <b>traditional attack</b> surface of the classic system supporting the ' + 'AI component. This is especially important in the T&E of the system as a whole, as the details of how the AI component is integrated with the traditional ' + 'system may expand or limit the entire attack surface, which might lead to AI corruption.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... An adversary ' + 'may alter the environment to cause the AI component to miss-classify an object, thus attacking the system without going through the <b>traditional attack</b> ' + 'surface. Another way an AI-enabled system may be attacked beyond the traditional vulnerabilities is by forcing the AI component to respond to many sensor ' + 'inputs by exposing some of the data used to train the AI model (known as an inversion attack against the ML system)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 1 more on page 108</i></div></td></tr>' + '</tbody></table>' + ''; } function x_operational_environments() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"operational environments"</b> shown in context from 4 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_operational_environments" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... Mitigating these risks may in volve new T&E approaches ' + 'that mimic an adversary\'s actions during all phases of the AI-component life cycle to assure that the AI component is resilient dur ing the development, ' + 'training, deployment, retraining, and adaptation of the AI implementation to different <b>operational environments</b>. Even with strong cyber defenses ' + 'in place, the adversary can still attack an air force AI-based system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... 5.2 GENERAL ' + 'RISKS OF AI-ENABLED SYSTEMS From a T&E perspective, the areas of an AI-enabled system that must be pro tected are the availability, integrity, and confidentiality ' + 'and privacy of the AI system: • Availability and integrity of the AI output • The integrity of the AI model, data, and software • Confidentiality ' + 'and privacy of the model and training data Several risks are inherent to using AI in <b>operational environments</b>. Awareness is the first step in mitigating ' + 'such risks, so the committee discusses here some main challenges that have been identified.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... While ' + 'this definition is relevant for AI-enabled systems, some additional vec tors should be considered in the life cycle of the AI component. For example, ' + 'an AI component relies on sensor data free from adversary manipulation, which may be difficult in <b>operational environments</b>.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... as it can generate a more specific set of <b>operational ' + 'environments</b> and constraints that is communicated to the operator during deployment. 5.8 RESEARCH IN T&E TO ADDRESS ADVERSARIAL AI To enable DAF T&E, ' + 'it will be important to distinguish between practical attacks by near-peer adversaries and academic attacks that would be impractical in deployed systems.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_software_component() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"software component"</b> shown in context from 4 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_software_component" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... This case is no different from any other <b>software ' + 'component</b> that must be tested, but the complexity of open-source AI solutions may be difficult to thoroughly test. Other types of analysis, such as ' + 'static and dynamic software testing, would be appropriate to augment the red team approach.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... Typically ' + 'this system is tuned and tested by a data scientist, part of the development team for the ML <b>software component</b>. One recent advance in defense of ' + 'ML components is using adversarial training.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 112...</div><div class=foundblock>... This step ' + 'is demonstrated in Figure 5-2. At the level of integration of the ML component into the operational system, the addition of <b>software component</b>s ' + 'and data paths expands the attack surface to include many elements that are likely exposed to the adversary, including the sensor input, data streams, ' + 'and specific APIs of the system to other systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=112\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... Robustness ' + 'is the property that a <b>software component</b> (that includes AI-enabled components) can meet mission requirements with variations in the operational ' + 'en vironment.7 Robustness is measured and tested during T&E with the introduction of deliberate perturbations of the environment beyond the initial configuration ' + 'and training of the <b>software component</b>.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_learning_model() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"learning model"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_learning_model" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such ' + 'as precision and recall, reduction of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, ' + 'such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against AI Systemsa Adversarial examples: These are inputs to a machine <b>learning ' + 'model</b> that are specifically designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created by adding ' + 'small, imperceptible perturbations to the input data designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Poisoning ' + 'attacks: These attacks involve introducing malicious or misleading data into a ma chine <b>learning model</b>\'s training data in an attempt to corrupt ' + 'the model\'s output. Evasion attacks: These attacks involve manipulating the input data specifically to evade detec tion by a machine <b>learning model</b> ' + 'that is being used for security or fraud detection purposes.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 2 more on page 106</i></div></td></tr>' + '</tbody></table>' + ''; } function x_inversion_attacks() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"inversion attacks"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_inversion_attacks" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Model <b>inversion attacks</b>: These attacks involve ' + 'attempting to recover sensitive information about the data used to train a machine learning model by manipulating the model\'s output and using it to infer ' + 'information about the input data. Also known as a data inference attack.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... AI model ' + '<b>inversion attacks</b> refer to techniques that aim to reverse engineer the internal workings of an AI model. These attacks are a type of adversarial ' + 'attack in which an attacker seeks to reconstruct the input data or features used to train a model or to generate synthetic inputs that will produce a desired ' + 'output from the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... Model ' + '<b>inversion attacks</b> are a potential concern because they could allow an attacker to learn sensitive information about the training data or the training ' + 'process itself, which could exploit the model\'s vulnerabilities or craft adversarial examples that can fool the model. Model <b>inversion attacks</b> ' + 'can be particularly dangerous when applied to models used in high-stakes situations, such as DoD </div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include model <b>inversion attacks</b> and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. ' + 'Even attacks that require unimpeded access to the AI model may be successful due to the transferability of adversarial attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_input_data() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"input data"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_input_data" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such ' + 'as precision and recall, reduction of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, ' + 'such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against AI Systemsa Adversarial examples: These are inputs to a machine learning ' + 'model that are specifically designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created by adding small, ' + 'imperceptible perturbations to the <b>input data</b> designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... Poisoning ' + 'attacks: These attacks involve introducing malicious or misleading data into a ma chine learning model\'s training data in an attempt to corrupt the ' + 'model\'s output. Evasion attacks: These attacks involve manipulating the <b>input data</b> specifically to evade detec tion by a machine learning model ' + 'that is being used for security or fraud detection purposes.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 1 more on page 106</i></div></td></tr><tr bgcolor=white><td ' + 'valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... AI model inversion attacks refer to techniques that aim to reverse engineer ' + 'the internal workings of an AI model. These attacks are a type of adversarial attack in which an attacker seeks to reconstruct the <b>input data</b> or ' + 'features used to train a model or to generate synthetic inputs that will produce a desired output from the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_dod_zero() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"dod zero"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_dod_zero" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Schneider, and N Carlini, 2022, \"<b>DoD Zero</b> ' + 'Trust Strat egy,\" Washington, DC: Department of Defense, https://dodcio.defense.gov/Portals/0/Documents/ Library/DoD-ZTStrategy.pdf.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... Finding 5-3: The impact of adversarial attacks on human-AI ' + 'enabled systems has not been well understood. At present, the <b>DoD Zero</b> Trust Strategy is only being implemented on en terprise systems.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... should ap ply the <b>DoD Zero</b> Trust Strategy to ' + 'all DAF artificial intelligence–enabled systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_cyberattack_surface() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"cyberattack surface"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_cyberattack_surface" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... The traditional <b>cyberattack surface</b> can be ' + 'considered a starting point for de fining an attack surface for an AI-enabled system. The rationale for starting with the <b>cyberattack surface</b> is ' + 'that the AI-enabled component is a data-driven software system, so it shares much of the same surface for an adversary to disrupt, deny, or degrade the ' + 'system containing the AI-enabled component, which could also pro vide access by the adversary to the data or software in the AI component.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... In addition to the <b>cyberattack surface</b>, the AI ' + 'component may enable an at tack surface beyond the traditional <b>cyberattack surface</b>. This access is due to the dependencies on data within the deployed ' + 'environment and in the backend infra structure, the supply chain of any AI model in the component, and the potential for retraining and adaptation with ' + 'an adversary-controlled environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_require_unimpeded() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"require unimpeded"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_require_unimpeded" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... is not based on network address or reusable tokens ' + 'but instead on the continuous authentication and authorization of principles of the access to datacenter services. Finding 5-1: Existing research on attacks ' + 'on AI-enabled systems and strate gies for mitigating them consider attacks that <b>require unimpeded</b> access to an underlying AI model.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... In particular, attacks that <b>require unimpeded</b> ' + 'access to an underlying AI model are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Nevertheless, ' + 'as stated 7 Variations regarding adversarial attacks were discussed in Section 3; in this section robustness also includes both adversarial and non-adversarial ' + 'variations in the environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. Even ' + 'attacks that <b>require unimpeded</b> access to the AI model may be successful due to the transferability of adversarial attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_adversarial_examples() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"adversarial examples"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_adversarial_examples" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... This may be in the form of statistical measures such ' + 'as precision and recall, reduction of a performance envelope required for a mission objective, or in a violation of the system\'s security requirements, ' + 'such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against AI Systemsa <b>Adversarial examples</b>: These are inputs to a machine ' + 'learning model that are specifically designed to mislead the model and cause it to make incorrect predictions. <b>Adversarial examples</b> can be created ' + 'by adding small, imperceptible perturbations to the input data designed to fool the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... Model ' + 'inversion attacks are a potential concern because they could allow an attacker to learn sensitive information about the training data or the training process ' + 'itself, which could exploit the model\'s vulnerabilities or craft <b>adversarial examples</b> that can fool the model. Model inversion attacks can be particularly ' + 'dangerous when applied to models used in high-stakes situations, such as DoD </div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... This augmented ' + 'dataset specifically attacks the function of the ML component, which can then be used as a training set in the component to drive down that spe cific ' + 'behavior. An iterative process of continuing to use <b>adversarial examples</b> and train the component to behave appropriately even in these conditions ' + 'can add to the robustness of the model.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_adversarial_manipulation() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"adversarial manipulation"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_adversarial_manipulation" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... , the data used for training and testing the component ' + 'is an important and critical element of exposure to potential <b>adversarial manipulation</b>. When open-source data are used, even if augmented with specific ' + 'mission data, the potential for <b>adversarial manipulation</b> of the open datasets to force specific ML behaviors can be difficult to detect and can ' + 'effectively compromise the ML-enabled system in deployment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... Protection ' + 'of the training and test datasets from <b>adversarial manipulation</b> is an important and specific need for ML components. Note that even exposing the ' + 'training and test datasets to the adversary without their manipulation can provide the adversary with an important tool to discover operational manipu ' + ' lation techniques that can force the ML-enabled system to fail during deployment, even if these components function properly in DT&E and OT&E.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... While not all aspects of AI-enabled systems may be amenable ' + 'to a ZT approach, the use of ZT where appropriate can decrease the risk of <b>adversarial manipulation</b> of the data or the software of the AI-enabled ' + 'system. Areas where this may not be applicable might be cases where the sensor network and data collection infrastructure is outside of the ZT boundary ' + 'for the system (e.g., open data for train ing purposes)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... Attacks ' + 'against the deployed system are also largely addressed by authentication and authorization of all elements, including sensors and other inputs to the AI-enabled ' + 'system, to prevent <b>adversarial manipulation</b> of any element in the pipeline of the AI-enabled system. Stages in securing the network to support operational ' + 'AI-enabled systems include: • Data flow mapping.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_data_sanitization() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"data sanitization"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_data_sanitization" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 119...</div><div class=foundblock>... We should also consider \"battle reserve\" models ' + 'as an approach. There are several promising areas of research that will improve the mitigation of adversarial AI including: • Techniques for <b>data ' + 'sanitization</b>.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=119\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 119...</div><div class=foundblock>... research ' + 'into new techniques for <b>data sanitization</b> is necessary to resolve this trade-off. The same research may be used in areas such as data privacy, where ' + 'specific personally identifiable information (PII)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=119\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_distribution_shifts() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"distribution shifts"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_distribution_shifts" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... For instance, having no training data from below-freezing ' + 'weather conditions means the AI model will not be accurate in such operational conditions. Additionally, AI implementations are very sensitive to <b>distribution ' + 'shifts</b>, which often happen gradually and slowly degrade the performance of an AI element, in dependent system, or joint cognitive system.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... . Even updates to sensor software can result in <b>distribution ' + 'shifts</b> in the data.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td><div class=foundblock>... <i>plus 1 more on page 104</i></div></td></tr>' + '</tbody></table>' + ''; } function x_test_datasets() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"test datasets"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_test_datasets" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... Thus, while this report focuses specifically on the ' + 'question of adversarial attacks against the AI components, it is important to con tinue to test and evaluate AI-enabled systems against a broad spectrum ' + 'of adversaries and attacks in a realistic operational context. In addition to development and <b>test datasets</b>, the collection and use of operational ' + 'data, especially in training AI compo nents, is key to mitigating threats to the AI component and the system as a whole.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... Protection ' + 'of the training and <b>test datasets</b> from adversarial manipulation is an important and specific need for ML components. Note that even exposing the ' + 'training and <b>test datasets</b> to the adversary without their manipulation can provide the adversary with an important tool to discover operational ' + 'manipu lation techniques that can force the ML-enabled system to fail during deployment, even if these components function properly in DT&E and OT&E.</div><div ' + 'class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... This will ensure that any network or component corruption ' + 'is contained in a single segment and does not spread to other networks and functions. Separating ML training and <b>test datasets</b> from operational ' + 'networks will reduce the attack surface for operational AI-enabled systems and mitigate several specific AI attacks.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_network_security() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"network security"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_network_security" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 112...</div><div class=foundblock>... In addition, <b>network security</b>, data protection, ' + 'encryp tion, and other classic defenses must be integrated at this point into the operational system and tested as part of the OT&E process. Finally, ' + 'the risk of adversarial attack during deployment is increased over the traditional software-intensive systems by creating an additional attack surface ' + ' FIGURE 5-2 The operational environment for ML-enabled systems.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=112\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... 5.6 <b>NETWORK ' + 'SECURITY</b> AND ZERO TRUST IMPLICATIONS Modern <b>network security</b> within DoD relies on the principles of zero trust (ZT)</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_unimpeded_access() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"unimpeded access"</b> shown in context from 3 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_unimpeded_access" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... is not based on network address or reusable tokens ' + 'but instead on the continuous authentication and authorization of principles of the access to datacenter services. Finding 5-1: Existing research on attacks ' + 'on AI-enabled systems and strate gies for mitigating them consider attacks that require <b>unimpeded access</b> to an underlying AI model.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read the entire page → </a></div></td></tr><tr ' + 'bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... In particular, attacks that require <b>unimpeded access</b> ' + 'to an underlying AI model are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Nevertheless, as ' + 'stated 7 Variations regarding adversarial attacks were discussed in Section 3; in this section robustness also includes both adversarial and non-adversarial ' + 'variations in the environment.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... These ' + 'may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. Even ' + 'attacks that require <b>unimpeded access</b> to the AI model may be successful due to the transferability of adversarial attacks.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_model_performance() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"model performance"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_model_performance" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... The current DAF requirements-driven T&E processes ' + 'can uncover the effect of these attacks if the AI <b>model performance</b> is degraded below the required, acceptable ranges. Thus DAF DT&E that tests ' + 'performance with hold-out datasets can be employed as a defense against some attacks.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... Recommendation ' + '5-1: The Department of the Air Force (DAF) should fund research activities that investigate the trade-offs between model resilience to adversarial attack ' + 'and <b>model performance</b> under operational conditions.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_attack_surfaces() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"attack surfaces"</b> shown in context from 2 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_attack_surfaces" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... . 5.4 <b>ATTACK SURFACES</b> FOR AI-ENABLED SYSTEMS ' + 'The attack surface of any system is defined by NIST (NIST SP 800-172 from GAO-19-128)</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107\');\">Read ' + 'the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... In summary, ' + 'the attack surface of AI-enabled systems encompasses all tra ditional <b>attack surfaces</b> inherent in software-intensive systems -- especially those ' + 'of ML-enabled systems -- but has the additional considerations of AI corruption through the life cycle of the AI system, including the data used in training, ' + 'test, and operations, and the access to the details in any configuration or model that is part of the AI component. 5.5 RISK OF ADVERSARIAL ATTACKS It ' + 'can be useful to divide the risk of adversarial attacks into the different levels of integration of the AI-enabled system.</div><div class=seepage><a href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109\');\">Read ' + 'the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } function x_deployed_environments() { document.getElementById("texttitle").innerHTML = '<table width=100% vspace=0 hspace=0 cellpadding=1 cellspacing=0><tbody><tr><td align=left><div class="active-filter"><b>"deployed environments"</b> shown in context from 1 pages. <a href="javascript:show_start();" class="clear-filter">x</a></div>' + '<img src="http://nap.nationalacademies.org/nap-cgi/webbug.cgi?referer=&dexterm_deployed_environments" height=1 width=1></td>' + '<td align=right></td></tr></tbody></table>'; document.getElementById("text2").innerHTML = p_i_c_table + '<tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... In addition, using specific APIs to control data ' + 'flow and access to the AI life cycle and <b>deployed environments</b> will limit the ability of the adversary to propagate attacks to multiple operational ' + 'systems. Note this is also a trade-off to a dynamic approach to DevSecOps that would enable near real-time updates to operational models based on retraining ' + 'and updates on changing <b>deployed environments</b> as well as the use of operational data in the development and training of new models.</div><div class=seepage><a ' + 'href=\"javascript:poe(\'http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114\');\">Read the entire page → </a></div></td></tr>' + '</tbody></table>' + ''; } // --> </script> <script> // Back To Top control $(function() { var offset = 220; var duration = 500; $(window).scroll(function() { if ($(this).scrollTop() > offset) { $('.back-to-top').fadeIn(duration); } else { $('.back-to-top').fadeOut(duration); } }); $('.back-to-top').on('click', function(event) { event.preventDefault(); $('html, body').animate({scrollTop: 0}, duration); return false; }); }); </script> <a name="top" id="top"></a> <div class="container"><br> <div class="row"><a href="//www.nap.edu/catalog.php?record_id=27092">Test and Evaluation Challenges in Artificial Intelligence-Enabled Systems for the Department of the Air Force </a> (2023) / <span class="text-muted">Chapter Skim</span> </div> <div class="heading row"> <!-- left column --> <div class="col-xs-10 col-sm-3 col-md-2 col-lg-2 sibcol1"> <!-- cover image --> <div class="col-xs-4 col-sm-10 col-md-10 col-lg-10"> <a href="https://www.nap.edu/catalog.php?record_id=27092"><img src="//www.nap.edu/cover/27092/200" class="img-responsive img-thumbnail" border="0"></a> </div> <!-- /cover image --> <!-- buying_info --> <div class="col-xs-6 col-sm-10 col-md-10 col-lg-10"> <!-- buying_info --> <div id="leftcolumn"> <div class="print"> <a href="http://nap.nationalacademies.org/cart/cart.php?record_id=27092&isbn=0-309-70439-1&action=buy it&fs=1"class="buyit btn btn-default"> <span data-icon="" class="icon-menu"> </span><span class="webprice">$42.00</span> <br> Buy PAPERBACK</a> </div> <div class="results-actions"><a href="http://nap.nationalacademies.org/openbook.php?record_id=27092" class="readit">Read Online</a></div><div class="results-actions"> <a href="http://nap.nationalacademies.org/download.php?record_id=27092" class="dloadit"><span>Download Free PDF</span></a> </div> </div></div> </div> <!-- end buying_info --> <!-- /left column --> <!-- center column --> <div class="col-xs-10 col-sm-7 col-md-8 col-lg-8 sibcol2"> <!-- Book title and author info --> <div class="row chaphits"><span class="text-muted">Currently Skimming:</span> <h4>5 AI Technical Risks Under Operational Conditions <br><span class="pgcount">Pages 102-120</span></h4></div> <div> <div style="padding: 5px 5px 5px 5px;"> <form method="get" action="/skim.php"> <input type="hidden" name="record_id" value="27092"> <div> <select name='chap' onChange="this.form.submit()"> <option value='' selected>Select Another Chapter</option> <option value="1-16">Summary, pages 1-16</option> <option value="17-34">1 Introduction, pages 17-34</option> <option value="35-48">2 Definitions and Perspectives, pages 35-48</option> <option value="49-78">3 Test and Evaluation of DAF AI-Enabled Systems, pages 49-78</option> <option value="79-101">4 Evolution of Test and Evaluation in Future AI-Based..., pages 79-101</option> <option value="102-120">5 AI Technical Risks Under Operational Conditions, pages 102-120</option> <option value="121-136">6 Emerging AI Technologies and Future T&E Implications, pages 121-136</option> <option value="137-138">7 Concluding Thoughts, pages 137-138</option> <option value="139-140">Appendixes, pages 139-140</option> <option value="141-141">Appendix A: Statement of Task, pages 141-141</option> <option value="142-148">Appendix B: Public Meeting Agendas, pages 142-148</option> <option value="149-155">Appendix C: Committee Member Biographical Information, pages 149-155</option> <option value="156-162">Appendix D: Acronyms and Abbreviations, pages 156-162</option> <option value="163-180">Appendix E: Testing, Evaluating, and Assessing Artificial..., pages 163-180</option> </select> <!--<span class="input-group-btn"> <button type="submit" class="btn btn-primary btn-sm"> SKIM </button> </span>--> </div> <input name="act" value="nap" type="hidden"></form> </div> </div> <div class="row"> <div class="col-xs-10 chaphit_desc"> <p class="small">The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter. <br> Select <b>key terms</b> on the right to highlight them within pages of the chapter. </p></div> <div class="col-xs-10"> <br> <div id="texttitle" class="texttitle"> </div> <div class="col-xs-7 col-md-8 chapterlist"> <div id="text2" class="text2"> <table width="97%" bgcolor=darkgray cellpadding=5 cellspacing=1 id="skimresults"><tbody> <tr bgcolor=white><td valign=top><div class=fromblock>From page 102...</div><div class=foundblock>... <font class="btext">The employment of AI-enabled systems can have significant benefits in augmenting the capabilities of the warfighter, but there are other risks inherent in the use of AI-enabled systems that must be considered. In particular, this chapter answers the second of the committee's three primary tasks, "consider examples of AI corruption under operational conditions and against malicious cyberattacks." 5.1 INTRODUCTION An AI-enabled system includes technical risks that must be considered during the test and evaluation (T&E)</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=102');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 103...</div><div class=foundblock>... <font class="btext">In addition, OT&E performance tests can ferret out operational environmental degrading ef fects, which may motivate model retraining using new datasets. Some attacks, referred to as backdoor attacks, only manifest once the adversary triggers them.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=103');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 104...</div><div class=foundblock>... <font class="btext">Instead, AI models learn to make predictions based on pat terns and relationships specific to the training data. As a result, they do not generalize well to novel data, and even a bit of extra signal noise in a sensor collection, which might be invisible to the human eye, can potentially confuse an AI system.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=104');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 105...</div><div class=foundblock>... <font class="btext">Thus, while this report focuses specifically on the question of adversarial attacks against the AI components, it is important to con tinue to test and evaluate AI-enabled systems against a broad spectrum of adversaries and attacks in a realistic operational context. In addition to development and test datasets, the collection and use of operational data, especially in training AI compo nents, is key to mitigating threats to the AI component and the system as a whole.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=105');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 106...</div><div class=foundblock>... <font class="btext">This may be in the form of statistical measures such as precision and recall, reduction of a performance envelope required for a mission objective, or in a violation of the system's security requirements, such as maintaining the secrecy BOX 5-1 Some Examples of Adversarial Attacks Against AI Systemsa Adversarial examples: These are inputs to a machine learning model that are specifically designed to mislead the model and cause it to make incorrect predictions. Adversarial examples can be created by adding small, imperceptible perturbations to the input data designed to fool the model.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=106');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 107...</div><div class=foundblock>... <font class="btext">Thus, all traditional testing for adversarial attacks against the attack surface that has been previously defined is still required. Defensive mechanisms that obvi ate or limit the capability of an adversary to take advantage of an attack surface are also the first line of defense for AI-enabled components.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=107');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 108...</div><div class=foundblock>... <font class="btext">AI model inversion attacks refer to techniques that aim to reverse engineer the internal workings of an AI model. These attacks are a type of adversarial attack in which an attacker seeks to reconstruct the input data or features used to train a model or to generate synthetic inputs that will produce a desired output from the model.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=108');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 109...</div><div class=foundblock>... <font class="btext">In summary, the attack surface of AI-enabled systems encompasses all tra ditional attack surfaces inherent in software-intensive systems -- especially those of ML-enabled systems -- but has the additional considerations of AI corruption through the life cycle of the AI system, including the data used in training, test, and operations, and the access to the details in any configuration or model that is part of the AI component. 5.5 RISK OF ADVERSARIAL ATTACKS It can be useful to divide the risk of adversarial attacks into the different levels of integration of the AI-enabled system.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=109');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 110...</div><div class=foundblock>... <font class="btext">Note, however, that over-training in the adversarial space could create vulnerabilities in the inversion attack of the resulting model, thus exposing which attacks and training sets are used to an adversary with limited access to the resulting deployed system. Recognizing the planned mission objectives is also important when testing these components.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=110');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 111...</div><div class=foundblock>... <font class="btext">Palat, and A Sinha, 2022, "Measuring AI Systems Beyond Accuracy," paper presented at AAAI Spring Symposium Series Workshop on AI Engineering: Creating Scalable, Human-Centered and Robust AI Systems, https://doi.org/10.48550/ arXiv.2204.04211.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=111');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 112...</div><div class=foundblock>... <font class="btext">In addition, network security, data protection, encryp tion, and other classic defenses must be integrated at this point into the operational system and tested as part of the OT&E process. Finally, the risk of adversarial attack during deployment is increased over the traditional software-intensive systems by creating an additional attack surface FIGURE 5-2 The operational environment for ML-enabled systems.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=112');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 113...</div><div class=foundblock>... <font class="btext">The execution of Zero Trust provides targeteers trusted, tagged, and labeled data so they can confi dently employ and share it with trusted partners, assured that the data is protected, secure, and accessed by only the people who need it when they need it, using least privilege principles.5 5 DoD, 2022, "Zero Trust Strategy." </font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=113');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 114...</div><div class=foundblock>... <font class="btext">In addition, using specific APIs to control data flow and access to the AI life cycle and deployed environments will limit the ability of the adversary to propagate attacks to multiple operational systems. Note this is also a trade-off to a dynamic approach to DevSecOps that would enable near real-time updates to operational models based on retraining and updates on changing deployed environments as well as the use of operational data in the development and training of new models.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=114');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 115...</div><div class=foundblock>... <font class="btext">Finding 5-3: The impact of adversarial attacks on human-AI enabled systems has not been well understood. At present, the DoD Zero Trust Strategy is only being implemented on en terprise systems.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=115');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 116...</div><div class=foundblock>... <font class="btext">The Department's ability to protect, and by extension, DoD personnel against the array of increasingly sophisticated cybersecurity threats depends on it. 5.7 ROBUST AND SECURE AI MODELS A common approach for increasing the robustness and security of AI models is the incorporation of monitoring or watchdog systems that compare the output of an AI-enabled system to pre-defined operational limits.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=116');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 117...</div><div class=foundblock>... <font class="btext">In particular, attacks that require unimpeded access to an underlying AI model are unlikely to be practical with traditional protections and mitigations inherent in deployed DAF systems. Nevertheless, as stated 7 Variations regarding adversarial attacks were discussed in Section 3; in this section robustness also includes both adversarial and non-adversarial variations in the environment.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=117');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 118...</div><div class=foundblock>... <font class="btext">These may include model inversion attacks and environmental manipulation attacks that exploit the mechanisms of the AI component without direct access. Even attacks that require unimpeded access to the AI model may be successful due to the transferability of adversarial attacks.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=118');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 119...</div><div class=foundblock>... <font class="btext">In many cases, sensitive data would need to be sanitized to prevent training data from exposure during model inver sion or when testing new types of AI-enabled systems. However, current approaches to sanitization do not effectively support the trade-off between the effective training of the system and the possibility of leaking sensitive data.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=119');">Read the entire page → </a></div></td></tr><tr bgcolor=white><td valign=top><div class=fromblock>From page 120...</div><div class=foundblock>... <font class="btext">120 Test and E va l u at i o n C h a l l e n g e s in AI - E n a b l e d S y s t e m s Conclusion 5-1: Promising areas of research that will improve the mitigation of adversarial AI include techniques for data sanitization, quantifiable uncertainty, and certifiable robustness. Additionally, Chapter 6 discusses additional emerging AI technologies and promising areas of research.</font> </div><div class=seepage><a href="javascript:poe('http://nap.nationalacademies.org/openbook.php?record_id=27092&page=120');">Read the entire page → </a></div></td></tr> </table> </div> </div> <div class="col-xs-3 col-md-2 pull-right termlist"> <p class="small">Key Terms</p> <li class="xy"><a href="javascript:x_adversarial_attacks();">adversarial attacks</a></li><li class="xy"><a href="javascript:x_zero_trust();">zero trust</a></li><li class="xy"><a href="javascript:x_training_data();">training data</a></li><li class="xy"><a href="javascript:x_operational_conditions();">operational conditions</a></li><li class="xy"><a href="javascript:x_operational_data();">operational data</a></li><div class=ider><br></div> <li class="xy"><a href="javascript:x_model_inversion();">model inversion</a></li><li class="xy"><a href="javascript:x_attacks_involve();">attacks involve</a></li><li class="xy"><a href="javascript:x_machine_learning();">machine learning</a></li><li class="xy"><a href="javascript:x_trust_strategy();">trust strategy</a></li><li class="xy"><a href="javascript:x_supply_chain();">supply chain</a></li><div class=ider><br></div> <li class="xy"><a href="javascript:x_adversarial_attack();">adversarial attack</a></li><li class="xy"><a href="javascript:x_traditional_attack();">traditional attack</a></li><li class="xy"><a href="javascript:x_operational_environments();">operational environments</a></li><li class="xy"><a href="javascript:x_software_component();">software component</a></li><li class="xy"><a href="javascript:x_learning_model();">learning model</a></li><div class=ider><br></div> <li class="xy"><a href="javascript:x_inversion_attacks();">inversion attacks</a></li><li class="xy"><a href="javascript:x_input_data();">input data</a></li><li class="xy"><a href="javascript:x_dod_zero();">dod zero</a></li><li class="xy"><a href="javascript:x_cyberattack_surface();">cyberattack surface</a></li><li class="xy"><a href="javascript:x_require_unimpeded();">require unimpeded</a></li><div class=ider><br></div> <li class="xy"><a href="javascript:x_adversarial_examples();">adversarial examples</a></li><li class="xy"><a href="javascript:x_adversarial_manipulation();">adversarial manipulation</a></li><li class="xy"><a href="javascript:x_data_sanitization();">data sanitization</a></li><li class="xy"><a href="javascript:x_distribution_shifts();">distribution shifts</a></li><li class="xy"><a href="javascript:x_test_datasets();">test datasets</a></li><div class=ider><br></div> <li class="xy"><a href="javascript:x_network_security();">network security</a></li><li class="xy"><a href="javascript:x_unimpeded_access();">unimpeded access</a></li><li class="xy"><a href="javascript:x_model_performance();">model performance</a></li><li class="xy"><a href="javascript:x_attack_surfaces();">attack surfaces</a></li><li class="xy"><a href="javascript:x_deployed_environments();">deployed environments</a></li> </div> <div class="col-xs-8 chapskim-nav"> <!-- use previous chapter phrase --> <div class="pull-left prev "><a href="/skim.php?record_id=27092&chap=79-101&act=nap" class="btn btn-default">← Previous Chapter Skim</a></div> <!-- use next chapter phrase --> <div class="pull-right next"><a href="/skim.php?record_id=27092&chap=121-136&act=nap" class="btn btn-default">Next Chapter Skim →</a></div> </div> <div class="clearfix"><br clear="all"></div><div class="clearfix"><br clear="all"></div> </div> <!-- / chapterlist and termlist--> <div class="col-xs-10" > <p class="small">This material may be derived from roughly machine-read images, and so is provided only to facilitate research.<br> <a href="//www.nap.edu/about/skimit.html">More information on Chapter Skim</a> is available.</p> <br><br> </div> </div> </div> </div> </div> <br><br> <!--Footer --> </body></html> <br> </div> <!--start NAP footer --> <footer class="global-footer" style="background: #211747; border-top: none" id="global-footer-nap"> <div class="container"> <div class="global-footer__orgs"> <div class="global-footer__logo"> <a href="https://nap.nationalacademies.org"><img src="/images/footer-nap-text.png" alt="National Academies Press: Washington, DC | Home" class="img-responsive"></a> </div> <div> <form id="subscribe-form" action="/updates/" method="POST"> <input type="hidden" name="action" value="subscribe"> <div class="input-group"> <label for="email" class="sr-only">Email Address</label> <input type="hidden" name="SOURCE" value="homepage"> <input class="subscribe-input form-control" type="text" name="EMAIL" id="email" placeholder="Sign up for e-mail updates" title="Enter your e-mail address to subscribe to updates from NAP."> <span class="subscribe-button"> <button type="submit" class="btn btn-subscribe">Subscribe</button> </span> </div> </form></div></div> <div class="global-footer__links"> <ul class="global-footer__nav"> <li> <h6>NAP Quick Links</h6> <ul> <li><a href="/">Home</a></li> <li><a href="/content/about-the-national-academies-press">About</a></li> <li><a href="https://notes.nap.edu/">Notes from NAP</a></li> </ul> <h6>Request A</h6> <ul> <li><a href="/content/request-desk-copies">Desk Copy</a> </li> <li><a href="/content/request/exam_copy.html">Exam Copy </a></li> <li><a href="/content/request-review-copies">Review Copy</a></li> <li><a href="/content/request-rights-and-permissions">Reprint Permission</a></li> </ul> </li> <li> <h6>Browse Publications</h6> <ul> <li><a href="/author">By Division</a></li> <li><a href="/new/">By New Releases</a></li> <li><a href="/collection">By Searchable Collections</a></li> <li><a href="/topic">By Topic</a></li> </ul> </li> <li> <h6>NAP Ordering</h6> <ul> <li><a href="/content/contact-customer-service">Contact Customer Service</a></li> </ul> <h6>Using NAP.edu</h6> <ul> <li><a href="/content/searching-nap-publications">Tips for Searching</a> </li> <li><a href="/content/using-mynap">Help With MyNAP</a></li> <li><a href="/content/about-pdfs">Downloading and Reading PDFs</a> </li> </ul> </li> <li> <h6>NAP Information</h6> <ul> <li><a href="/content/help-with-discounts">Discounts</a></li> <li><a href="/content/information-for-book-trade-customers">Book Trade Customers</a></li> <li><a href="/content/request-rights-and-permissions">Rights and Permissions</a></li> <li><a href="/content/international-ordering-and-distribution">International Ordering and Distribution</a></li> <li><a href="/content/translation-rights">Translation Rights</a></li> </ul></li> </ul> </div></div></footer> <!--start NASEM footer --> <footer class="global-footer" style="background: #666; border-top: none"><div class="container"><div class="global-footer__orgs"><div class="global-footer__logo"> <a aria-label="Logo of The National Academies" href="https://www.nationalacademies.org/home"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 315.11 58.33"><defs><style>.cls-1{fill:#fff;}</style></defs><g id="Layer_2" data-name="Layer 2"><g id="Layer_1-2" data-name="Layer 1"><polygon class="cls-1" points="3.73 7.55 20.84 25.49 20.84 1.5 17.81 1.5 17.81 18.08 0.7 0.14 0.7 24.06 3.73 24.06 3.73 7.55"/><polygon class="cls-1" points="28.49 24.06 31.84 24.06 38.98 7.25 45.92 23.87 45.99 24.06 49.34 24.06 39.05 0 28.49 24.06"/><polygon class="cls-1" points="153.23 24.06 156.58 24.06 163.72 7.25 170.66 23.87 170.73 24.06 174.08 24.06 163.79 0 153.23 24.06"/><polygon class="cls-1" points="59.72 24.06 62.76 24.06 62.76 4.41 69.52 4.41 69.52 1.5 52.96 1.5 52.96 4.41 59.72 4.41 59.72 24.06"/><rect class="cls-1" x="78.85" y="1.5" width="3.03" height="22.55"/><path class="cls-1" d="M115.69,12.8a11.72,11.72,0,0,0-23.44,0,11.72,11.72,0,0,0,23.44,0m-3.09,0A8.63,8.63,0,1,1,104,4a8.7,8.7,0,0,1,8.63,8.76"/><polygon class="cls-1" points="128.29 7.55 145.4 25.49 145.4 1.5 142.37 1.5 142.37 18.08 125.25 0.14 125.25 24.06 128.29 24.06 128.29 7.55"/><polygon class="cls-1" points="184.65 1.5 181.62 1.5 181.62 24.06 193.7 24.06 193.7 21.16 184.65 21.16 184.65 1.5"/><path class="cls-1" d="M37.15,34.08A11.48,11.48,0,0,0,25.37,45.8,11.36,11.36,0,0,0,37,57.38a11.77,11.77,0,0,0,6.79-2.19l.13-.09V51.2l-.51.51A9.77,9.77,0,0,1,41,53.5a8.9,8.9,0,0,1-3.95,1A8.75,8.75,0,0,1,37,37a9.09,9.09,0,0,1,3.66.82,10.71,10.71,0,0,1,2.78,2l.51.5V36.4l-.12-.09a10.62,10.62,0,0,0-6.67-2.23"/><path class="cls-1" d="M91.91,37.08a12.19,12.19,0,0,0-8.29-2.64H78.85V57h4.74c3.72,0,6-.73,8.22-2.61a10.75,10.75,0,0,0,4-8.65,11,11,0,0,0-3.86-8.65m-2.09,15.1a9.11,9.11,0,0,1-6.1,1.91H81.88V37.34h1.84c5.44,0,9,3.29,9,8.39a8.09,8.09,0,0,1-2.85,6.45"/><polygon class="cls-1" points="104.99 56.99 118.4 56.99 118.4 54.09 108.02 54.09 108.02 46.16 117.11 46.16 117.11 43.26 108.02 43.26 108.02 37.34 118.4 37.34 118.4 34.44 104.99 34.44 104.99 56.99"/><polygon class="cls-1" points="176.17 56.99 189.58 56.99 189.58 54.09 179.2 54.09 179.2 46.16 188.28 46.16 188.28 43.26 179.2 43.26 179.2 37.34 189.58 37.34 189.58 34.44 176.17 34.44 176.17 56.99"/><polygon class="cls-1" points="138.88 51.34 130.31 32.49 126.44 56.99 129.55 56.99 131.62 42.47 138.84 58.33 146.15 42.47 148.13 56.99 151.23 56.99 147.61 32.45 138.88 51.34"/><rect class="cls-1" x="161.31" y="34.44" width="3.03" height="22.55"/><path class="cls-1" d="M209,46.15A10.78,10.78,0,0,0,205.49,44l-2.12-.92a6.8,6.8,0,0,1-2.21-1.33,2,2,0,0,1-.73-1.69A3.07,3.07,0,0,1,203.82,37a3.62,3.62,0,0,1,2.28.69,4.87,4.87,0,0,1,1.2,1.55l.15.3,2.45-1.62-.12-.24a7,7,0,0,0-2.24-2.55,6.34,6.34,0,0,0-3.69-1c-3.8,0-6.45,2.45-6.45,6a4.86,4.86,0,0,0,1.6,3.8,9.51,9.51,0,0,0,2.84,1.78l2.15,1a9.5,9.5,0,0,1,2.56,1.46,3,3,0,0,1,1,2.35,4.2,4.2,0,0,1-7.06,2.92,4.93,4.93,0,0,1-1.24-2.64l-.05-.33-3,.82.06.28a8.32,8.32,0,0,0,2.13,4,6.74,6.74,0,0,0,5,1.87,6.91,6.91,0,0,0,7.24-7A5.58,5.58,0,0,0,209,46.15"/><polygon class="cls-1" points="0 56.99 3.35 56.99 10.49 40.18 17.43 56.81 17.5 56.99 20.86 56.99 10.56 32.93 0 56.99"/><polygon class="cls-1" points="50.21 56.99 53.57 56.99 60.71 40.18 67.64 56.81 67.72 56.99 71.07 56.99 60.78 32.93 50.21 56.99"/><path class="cls-1" d="M233.74,2.1a3.34,3.34,0,0,0-.93-.11,2.53,2.53,0,0,0-1.52.45,1.48,1.48,0,0,0-.63,1.28,2,2,0,0,0,.54,1.44,7.94,7.94,0,0,0,1.64,1.26,13.5,13.5,0,0,1,1.26.89,3.76,3.76,0,0,1,.83,1,2.62,2.62,0,0,1,.33,1.33,3.06,3.06,0,0,1-.5,1.74,3.26,3.26,0,0,1-1.4,1.18,5.06,5.06,0,0,1-2.09.41,8.3,8.3,0,0,1-1.94-.22,3.51,3.51,0,0,1-1.24-.46l.34-2.18h.92l.09,1.62a2.53,2.53,0,0,0,.81.33,3.86,3.86,0,0,0,1,.13,2.93,2.93,0,0,0,1.77-.53,1.68,1.68,0,0,0,.73-1.47,1.91,1.91,0,0,0-.49-1.34,6.42,6.42,0,0,0-1.49-1.16c-.52-.34-1-.65-1.32-.94a3.76,3.76,0,0,1-.9-1,2.64,2.64,0,0,1-.37-1.4,2.64,2.64,0,0,1,.5-1.62,3.23,3.23,0,0,1,1.35-1.08,4.58,4.58,0,0,1,1.86-.37,7,7,0,0,1,1.72.18,3.31,3.31,0,0,1,1.09.45l-.35,2h-.83l-.12-1.47a2.39,2.39,0,0,0-.68-.3"/><path class="cls-1" d="M241,5a2,2,0,0,0-1.35.54,3.66,3.66,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,1.95,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a3.84,3.84,0,0,1-1.14.95,3.43,3.43,0,0,1-1.76.43,2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54A6.43,6.43,0,0,1,237.34,7a5.3,5.3,0,0,1,1.55-2.08,3.74,3.74,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3.09,3.09,0,0,1,.81.26l-.42,1.49-.26-.09A2.53,2.53,0,0,0,241,5"/><path class="cls-1" d="M246.1,11.61a.56.56,0,0,0,.07.29.21.21,0,0,0,.18.1.65.65,0,0,0,.34-.11,2.64,2.64,0,0,0,.44-.35l.2.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.31-2.16.52-3.7l-1-.35.09-.52,2.18-.36.34.17-1,7.22ZM246,2.15a.84.84,0,0,1-.23-.59,1.13,1.13,0,0,1,.3-.79.94.94,0,0,1,.71-.32.79.79,0,0,1,.61.24.77.77,0,0,1,.23.57,1.15,1.15,0,0,1-.29.81,1,1,0,0,1-.77.33.75.75,0,0,1-.56-.25"/><path class="cls-1" d="M254,12.43a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.72-.47,2.74,2.74,0,0,1-1-1.23A4,4,0,0,1,249,9.62a6.35,6.35,0,0,1,.59-2.75,5.08,5.08,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.27,2.27,0,0,1,1.64.52A1.79,1.79,0,0,1,255.72,6a2.34,2.34,0,0,1-.89,1.84,5.79,5.79,0,0,1-2.08,1.12,8.1,8.1,0,0,1-2.15.4,3.42,3.42,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57A4,4,0,0,1,254,12.43m-2.11-7A4.1,4.1,0,0,0,251,6.88a5.9,5.9,0,0,0-.39,1.69,5.31,5.31,0,0,0,1.62-.33,4,4,0,0,0,1.41-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-1" d="M261.24,4.62A3.14,3.14,0,0,1,263,4.06a1.67,1.67,0,0,1,1.32.52,2.61,2.61,0,0,1,.46,1.71,15.44,15.44,0,0,1-.34,2.34l-.06.36-.13.74a11.51,11.51,0,0,0-.24,1.85c0,.28.06.42.2.42a1.22,1.22,0,0,0,.72-.39l.18.4a1.54,1.54,0,0,1-.32.36,2.92,2.92,0,0,1-.61.39,1.73,1.73,0,0,1-.79.17.84.84,0,0,1-1-.93,12.88,12.88,0,0,1,.27-1.88l.14-.73.06-.37c.1-.55.19-1,.25-1.45a8.17,8.17,0,0,0,.09-1.16,1.67,1.67,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.33,2.33,0,0,0-1.37.61,5.75,5.75,0,0,0-1.3,1.47l-.75,5.54h-1.46l1-7.43L257.62,5l.12-.58,1.88-.34.32.17-.2,1.84a5.72,5.72,0,0,1,1.5-1.45"/><path class="cls-1" d="M270.77,5a2,2,0,0,0-1.35.54,3.66,3.66,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,1.95,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a3.84,3.84,0,0,1-1.14.95,3.43,3.43,0,0,1-1.76.43,2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54A6.43,6.43,0,0,1,267.13,7a5.3,5.3,0,0,1,1.55-2.08,3.76,3.76,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3,3,0,0,1,.81.26l-.42,1.49-.26-.09A2.53,2.53,0,0,0,270.77,5"/><path class="cls-1" d="M278.66,12.43a3.75,3.75,0,0,1-2,.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63,6.35,6.35,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52A1.82,1.82,0,0,1,280.38,6a2.34,2.34,0,0,1-.89,1.84,5.79,5.79,0,0,1-2.08,1.12,8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87A1.48,1.48,0,0,0,277,12a3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-1" d="M286.18,5.13a2.75,2.75,0,0,0-1.09-.21,1.85,1.85,0,0,0-1.13.31,1,1,0,0,0-.43.89,1.43,1.43,0,0,0,.34.93,6.08,6.08,0,0,0,1,.88,10.43,10.43,0,0,1,.88.74,3.1,3.1,0,0,1,.6.76,2.15,2.15,0,0,1,.26,1.05,2.09,2.09,0,0,1-.9,1.81,3.71,3.71,0,0,1-2.22.64,6.32,6.32,0,0,1-1.37-.15,2.1,2.1,0,0,1-.9-.34l.24-1.06a3.41,3.41,0,0,0,.88.44,3.37,3.37,0,0,0,1.18.21,2.21,2.21,0,0,0,1.15-.29,1,1,0,0,0,.48-.92,1.37,1.37,0,0,0-.38-1,6.83,6.83,0,0,0-1.1-.89,3.07,3.07,0,0,1-1.58-2.5,2.16,2.16,0,0,1,.4-1.27,2.64,2.64,0,0,1,1.09-.84,3.81,3.81,0,0,1,1.54-.3,6.32,6.32,0,0,1,1.13.1,4.07,4.07,0,0,1,.8.19l-.26,1.17a1.55,1.55,0,0,0-.6-.39"/><polygon class="cls-1" points="229.34 23.43 236.61 23.43 236.51 25.87 235.6 25.87 235.35 24.25 231.87 24.13 231.29 28.48 234.85 28.36 234.74 29.51 231.21 29.38 230.58 34.05 234.91 33.87 235.77 32.01 236.63 32.18 236.01 34.84 227.85 34.84 227.91 34.24 229.06 34.05 230.29 24.25 229.24 24.09 229.34 23.43"/><path class="cls-1" d="M242.18,26.67a3.16,3.16,0,0,1,1.73-.56,1.67,1.67,0,0,1,1.32.53,2.51,2.51,0,0,1,.46,1.7,14.42,14.42,0,0,1-.34,2.35l-.06.35-.12.74a10.81,10.81,0,0,0-.25,1.86c0,.27.07.41.2.41a1.15,1.15,0,0,0,.72-.38l.19.4a1.54,1.54,0,0,1-.32.35,2.46,2.46,0,0,1-.61.39,1.76,1.76,0,0,1-.79.18c-.65,0-1-.31-1-.94a11.94,11.94,0,0,1,.28-1.87l.14-.74.06-.37c.1-.55.18-1,.24-1.45a8.14,8.14,0,0,0,.1-1.16,1.64,1.64,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.3,2.3,0,0,0-1.38.61,5.67,5.67,0,0,0-1.29,1.46l-.76,5.55h-1.45l1-7.44-.76-.36.13-.59,1.87-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-1" d="M252.77,37.29a4,4,0,0,1-1.55,1.05,5.36,5.36,0,0,1-1.78.32,5.69,5.69,0,0,1-1-.09,3.45,3.45,0,0,1-.76-.19l.32-1.24a5,5,0,0,0,.9.34,3.56,3.56,0,0,0,1.06.17,2.22,2.22,0,0,0,1.45-.39,2.12,2.12,0,0,0,.63-1.41l.29-2a3.73,3.73,0,0,1-1,.84A2.61,2.61,0,0,1,250,35a2.16,2.16,0,0,1-1.89-.88,3.9,3.9,0,0,1-.64-2.35,6.52,6.52,0,0,1,.58-2.6A5.62,5.62,0,0,1,249.73,27a4.15,4.15,0,0,1,2.61-.87,4.47,4.47,0,0,1,.87.1,5.15,5.15,0,0,1,.8.24l1-.34-1.38,9.31a3.37,3.37,0,0,1-.81,1.87m-.61-10.36a2.36,2.36,0,0,0-1.75.74,4.82,4.82,0,0,0-1.07,1.85,7.16,7.16,0,0,0-.35,2.15,3.3,3.3,0,0,0,.37,1.72,1.16,1.16,0,0,0,1,.62,2.09,2.09,0,0,0,1.2-.39,3.12,3.12,0,0,0,.9-.87l.82-5.5a1.85,1.85,0,0,0-1.15-.32"/><path class="cls-1" d="M258.39,33.67a.45.45,0,0,0,.07.28.19.19,0,0,0,.17.1.6.6,0,0,0,.35-.11,3.13,3.13,0,0,0,.44-.35l.2.37a3.17,3.17,0,0,1-.69.68,1.77,1.77,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.24.82.82,0,0,1-.22-.71c0-.26.16-1.25.4-3,.15-.93.32-2.17.52-3.7l-.95-.36.09-.52,2.19-.35.33.17-1,7.21Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.09,1.09,0,0,1,.3-.78.94.94,0,0,1,.72-.33.81.81,0,0,1,.61.24.82.82,0,0,1,.23.58,1.17,1.17,0,0,1-.29.8,1,1,0,0,1-.77.33.76.76,0,0,1-.56-.24"/><path class="cls-1" d="M265.74,26.67a3.11,3.11,0,0,1,1.72-.56,1.65,1.65,0,0,1,1.32.53,2.5,2.5,0,0,1,.47,1.7,15.68,15.68,0,0,1-.34,2.35l-.06.35-.13.74a11.6,11.6,0,0,0-.24,1.86c0,.27.06.41.2.41a1.2,1.2,0,0,0,.72-.38l.18.4a1.54,1.54,0,0,1-.32.35,2.28,2.28,0,0,1-.61.39,1.73,1.73,0,0,1-.79.18.84.84,0,0,1-1-.94,12.8,12.8,0,0,1,.27-1.87l.14-.74.06-.37c.1-.55.19-1,.25-1.45a8.07,8.07,0,0,0,.09-1.16,1.64,1.64,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.27,2.27,0,0,0-1.37.61,5.7,5.7,0,0,0-1.3,1.46l-.75,5.55h-1.46l1-7.44-.75-.36.12-.59,1.88-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-1" d="M276.11,34.48a3.7,3.7,0,0,1-1.95.52,3,3,0,0,1-1.71-.46,2.87,2.87,0,0,1-1-1.23,4.24,4.24,0,0,1-.32-1.64,6.53,6.53,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.76,3.76,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.53,1.76,1.76,0,0,1,.55,1.36,2.31,2.31,0,0,1-.88,1.84A5.83,5.83,0,0,1,274.87,31a9.08,9.08,0,0,1-2.15.4,3.49,3.49,0,0,0,.36,1.86,1.48,1.48,0,0,0,1.4.79,3.16,3.16,0,0,0,1.35-.31,3.9,3.9,0,0,0,1.22-.9l.33.57a3.93,3.93,0,0,1-1.27,1.1m-2.1-7a4.12,4.12,0,0,0-.88,1.44,6.37,6.37,0,0,0-.38,1.7,5.17,5.17,0,0,0,1.62-.34,3.78,3.78,0,0,0,1.4-.87,1.74,1.74,0,0,0,.57-1.28,1.31,1.31,0,0,0-.28-.91,1,1,0,0,0-.78-.32,1.68,1.68,0,0,0-1.27.58"/><path class="cls-1" d="M284,34.48a3.7,3.7,0,0,1-2,.52,3,3,0,0,1-1.71-.46,2.87,2.87,0,0,1-1-1.23,4.24,4.24,0,0,1-.32-1.64,6.53,6.53,0,0,1,.59-2.75,5.26,5.26,0,0,1,1.63-2,3.79,3.79,0,0,1,2.3-.77,2.31,2.31,0,0,1,1.64.53,1.76,1.76,0,0,1,.55,1.36,2.31,2.31,0,0,1-.88,1.84A5.74,5.74,0,0,1,282.73,31a9.08,9.08,0,0,1-2.15.4,3.4,3.4,0,0,0,.37,1.86,1.45,1.45,0,0,0,1.39.79,3.2,3.2,0,0,0,1.36-.31,3.86,3.86,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.28,1.1m-2.1-7a4.12,4.12,0,0,0-.88,1.44,6.37,6.37,0,0,0-.38,1.7,5.17,5.17,0,0,0,1.62-.34,3.7,3.7,0,0,0,1.4-.87,1.74,1.74,0,0,0,.57-1.28,1.31,1.31,0,0,0-.28-.91,1,1,0,0,0-.78-.32,1.7,1.7,0,0,0-1.27.58"/><path class="cls-1" d="M290.05,27.3a3.77,3.77,0,0,1,1-.83,2.34,2.34,0,0,1,1.22-.34.83.83,0,0,1,.5.11l-.33,1.62a1.45,1.45,0,0,0-.75-.15,2.3,2.3,0,0,0-1.29.44,2.89,2.89,0,0,0-1,1.22l-.7,5.47h-1.48l1-7.41-.77-.38.11-.57,1.81-.35.31.18-.11,1.37-.08.61a4.09,4.09,0,0,1,.6-1"/><path class="cls-1" d="M295.29,33.67a.45.45,0,0,0,.07.28.19.19,0,0,0,.17.1.6.6,0,0,0,.35-.11,3.13,3.13,0,0,0,.44-.35l.2.37a3.17,3.17,0,0,1-.69.68,1.77,1.77,0,0,1-1.08.35,1.14,1.14,0,0,1-.78-.24.82.82,0,0,1-.22-.71c0-.26.16-1.25.4-3,.15-.93.32-2.17.52-3.7l-.95-.36.09-.52,2.19-.35.33.17-1,7.21Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.09,1.09,0,0,1,.3-.78.94.94,0,0,1,.72-.33.81.81,0,0,1,.61.24.82.82,0,0,1,.23.58,1.17,1.17,0,0,1-.29.8,1,1,0,0,1-.77.33.76.76,0,0,1-.56-.24"/><path class="cls-1" d="M302.33,26.67a3.14,3.14,0,0,1,1.73-.56,1.67,1.67,0,0,1,1.32.53,2.56,2.56,0,0,1,.46,1.7,15.68,15.68,0,0,1-.34,2.35l-.06.35-.12.74a10.81,10.81,0,0,0-.25,1.86c0,.27.07.41.2.41a1.2,1.2,0,0,0,.72-.38l.19.4a1.64,1.64,0,0,1-.33.35,2.22,2.22,0,0,1-.6.39,1.76,1.76,0,0,1-.79.18c-.65,0-1-.31-1-.94a12.8,12.8,0,0,1,.27-1.87l.14-.74.06-.37c.11-.55.19-1,.25-1.45a8.07,8.07,0,0,0,.09-1.16,1.73,1.73,0,0,0-.2-1,.88.88,0,0,0-.73-.27,2.3,2.3,0,0,0-1.38.61,5.7,5.7,0,0,0-1.3,1.46l-.75,5.55h-1.46l1-7.44-.75-.36.13-.59,1.87-.34.32.17-.2,1.85a5.77,5.77,0,0,1,1.5-1.46"/><path class="cls-1" d="M312.92,37.29a4,4,0,0,1-1.55,1.05,5.36,5.36,0,0,1-1.78.32,5.69,5.69,0,0,1-1-.09,3.57,3.57,0,0,1-.77-.19l.33-1.24a5,5,0,0,0,.9.34,3.51,3.51,0,0,0,1.06.17,2.22,2.22,0,0,0,1.45-.39,2.11,2.11,0,0,0,.62-1.41l.3-2a3.73,3.73,0,0,1-1,.84,2.63,2.63,0,0,1-1.33.34,2.16,2.16,0,0,1-1.89-.88,3.9,3.9,0,0,1-.64-2.35,6.35,6.35,0,0,1,.58-2.6A5.62,5.62,0,0,1,309.88,27a4.11,4.11,0,0,1,2.61-.87,4.56,4.56,0,0,1,.87.1,5.15,5.15,0,0,1,.8.24l.95-.34-1.38,9.31a3.37,3.37,0,0,1-.81,1.87m-.62-10.36a2.35,2.35,0,0,0-1.74.74,4.7,4.7,0,0,0-1.07,1.85,7.16,7.16,0,0,0-.35,2.15,3.3,3.3,0,0,0,.37,1.72,1.14,1.14,0,0,0,1,.62,2,2,0,0,0,1.19-.39,3.3,3.3,0,0,0,.91-.87l.82-5.5a1.88,1.88,0,0,0-1.16-.32"/><path class="cls-1" d="M237.11,49.24c.83-1.82,1.39-3.07,1.65-3.76h3.29l-.06.66-1.37.17-.29,9.8,1.31.18-.07.6h-4.22l.06-.6,1.4-.18.35-6.79.2-3.12L238.2,49l-3.57,8-.83.11-1.26-7.89-.47-2.92-.63,3-1.64,6.91,1.55.18-.06.6h-4l.06-.6,1.28-.18,2.47-9.8-1.41-.17.09-.66h3.56L234.56,53l.2,1.57.67-1.53c.28-.69.84-1.95,1.68-3.77"/><path class="cls-1" d="M247.75,56.54a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63,6.35,6.35,0,0,1,.59-2.75,5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52,1.82,1.82,0,0,1,.55,1.37,2.34,2.34,0,0,1-.89,1.84A5.79,5.79,0,0,1,246.5,53a8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87A1.76,1.76,0,0,0,248,50.2a1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/><path class="cls-1" d="M257.17,55.72q0,.39.24.39a1.19,1.19,0,0,0,.7-.39l.22.39a3.42,3.42,0,0,1-.67.62,1.68,1.68,0,0,1-1,.31c-.6,0-.9-.32-.9-1v-.34a3.94,3.94,0,0,1-1.08.94,2.72,2.72,0,0,1-1.44.38,2.13,2.13,0,0,1-1.87-.88,4,4,0,0,1-.63-2.34,6.41,6.41,0,0,1,.58-2.62A5.56,5.56,0,0,1,253,49a4.1,4.1,0,0,1,2.61-.86,4,4,0,0,1,1.15.18l.41-2.76-1-.26.13-.54,2.16-.28.34.2-1.6,10.84a.79.79,0,0,0,0,.17M255.42,49a2.32,2.32,0,0,0-1.74.75,4.64,4.64,0,0,0-1.06,1.84,7.15,7.15,0,0,0-.34,2.16,3.44,3.44,0,0,0,.36,1.72,1.12,1.12,0,0,0,1,.61,2.06,2.06,0,0,0,1.23-.41,3.11,3.11,0,0,0,.9-.91l.83-5.42a1.94,1.94,0,0,0-1.2-.34"/><path class="cls-1" d="M262.14,55.72a.57.57,0,0,0,.06.29.24.24,0,0,0,.18.1.71.71,0,0,0,.35-.11,3.84,3.84,0,0,0,.44-.35l.19.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.19,1.19,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.32-2.16.52-3.7l-.95-.35.09-.52,2.18-.36.34.17-1,7.22Zm-.07-9.46a.81.81,0,0,1-.24-.6,1.12,1.12,0,0,1,.3-.78,1,1,0,0,1,.71-.32.83.83,0,0,1,.62.24.81.81,0,0,1,.23.57,1.2,1.2,0,0,1-.29.81,1,1,0,0,1-.77.33.73.73,0,0,1-.56-.25"/><path class="cls-1" d="M269.42,49.14a2,2,0,0,0-1.35.54,3.6,3.6,0,0,0-.92,1.52,7,7,0,0,0-.33,2.24,3.52,3.52,0,0,0,.43,2,1.39,1.39,0,0,0,1.24.66,2.38,2.38,0,0,0,1.1-.24,5.05,5.05,0,0,0,1-.72l.3.57a4,4,0,0,1-1.14,1A3.43,3.43,0,0,1,268,57a2.44,2.44,0,0,1-2-.91,4,4,0,0,1-.73-2.54,6.43,6.43,0,0,1,.53-2.51A5.3,5.3,0,0,1,267.33,49a3.69,3.69,0,0,1,2.42-.83,5,5,0,0,1,.93.09,3.09,3.09,0,0,1,.81.26L271.07,50l-.26-.09a2.53,2.53,0,0,0-1.39-.78"/><path class="cls-1" d="M274.77,55.72a.56.56,0,0,0,.07.29.24.24,0,0,0,.18.1.71.71,0,0,0,.35-.11,3.15,3.15,0,0,0,.43-.35l.2.37a2.88,2.88,0,0,1-.68.67,1.71,1.71,0,0,1-1.08.35,1.19,1.19,0,0,1-.78-.23.8.8,0,0,1-.22-.72c0-.25.15-1.25.4-3,.14-.93.32-2.16.52-3.7l-1-.35.09-.52,2.18-.36.34.17-1,7.22Zm-.07-9.46a.85.85,0,0,1-.23-.6,1.12,1.12,0,0,1,.3-.78.94.94,0,0,1,.71-.32.79.79,0,0,1,.61.24.78.78,0,0,1,.24.57,1.16,1.16,0,0,1-.3.81,1,1,0,0,1-.76.33.75.75,0,0,1-.57-.25"/><path class="cls-1" d="M282,48.73a3.09,3.09,0,0,1,1.73-.56,1.65,1.65,0,0,1,1.32.52,2.54,2.54,0,0,1,.46,1.7,15.09,15.09,0,0,1-.33,2.35l-.07.36-.12.73a11.6,11.6,0,0,0-.24,1.86c0,.28.06.42.19.42a1.22,1.22,0,0,0,.73-.39l.18.4a1.77,1.77,0,0,1-.32.36,3.21,3.21,0,0,1-.61.39,1.86,1.86,0,0,1-.79.17c-.64,0-1-.31-1-.93a13.42,13.42,0,0,1,.28-1.88l.14-.73.06-.37c.1-.56.18-1,.25-1.45a8.17,8.17,0,0,0,.09-1.16,1.67,1.67,0,0,0-.21-1,.87.87,0,0,0-.73-.27,2.31,2.31,0,0,0-1.37.61,5.58,5.58,0,0,0-1.3,1.47l-.75,5.54H278.2l1-7.43-.75-.37.12-.58,1.87-.34.33.17-.2,1.84A5.85,5.85,0,0,1,282,48.73"/><path class="cls-1" d="M292.42,56.54a3.75,3.75,0,0,1-1.95.52,3,3,0,0,1-1.71-.47,2.7,2.7,0,0,1-1-1.23,4.19,4.19,0,0,1-.32-1.63A6.35,6.35,0,0,1,288,51a5.17,5.17,0,0,1,1.63-2,3.7,3.7,0,0,1,2.29-.77,2.32,2.32,0,0,1,1.65.52,1.82,1.82,0,0,1,.55,1.37,2.34,2.34,0,0,1-.89,1.84A5.79,5.79,0,0,1,291.17,53a8,8,0,0,1-2.15.4,3.51,3.51,0,0,0,.37,1.87,1.48,1.48,0,0,0,1.4.79,3.18,3.18,0,0,0,1.35-.32,3.75,3.75,0,0,0,1.21-.9l.34.57a4,4,0,0,1-1.27,1.11m-2.11-7a4.25,4.25,0,0,0-.87,1.45,6.25,6.25,0,0,0-.39,1.69,5.44,5.44,0,0,0,1.63-.33,4.05,4.05,0,0,0,1.4-.87,1.76,1.76,0,0,0,.57-1.28,1.33,1.33,0,0,0-.29-.92,1,1,0,0,0-.77-.31,1.74,1.74,0,0,0-1.28.57"/></g></g></svg> </a> </div> <div><a href="http://www.nasonline.org/" rel="noreferrer" target="_blank" class="global-footer__org"><img src="data:image/png;base64,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" alt="NAS Logo" width="17" height="17"> <span> National Academy of Sciences </span></a> <a href="https://www.nae.edu/" rel="noreferrer" target="_blank" class="global-footer__org"><img src="data:image/png;base64,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" alt="NAE Logo" width="17" height="17"> <span> National Academy of Engineering </span></a> <a href="https://nam.edu/" rel="noreferrer" target="_blank" class="global-footer__org"><img src="data:image/png;base64,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" alt="NAM Logo" width="17" height="17"> <span> National Academy of Medicine </span></a></div></div> <div class="global-footer__links"><ul class="global-footer__nav"><li><a rel="" title="" href="//www.nationalacademies.org/about" target="_self"> About Us </a></li><li><a rel="" title="" href="//www.nationalacademies.org/our-work" target="_self"> Our Work </a></li><li><a rel="" title="" href="//www.nationalacademies.org/events" target="_self"> Events </a></li><li><a rel="" title="" href="//www.nationalacademies.org/contact-us" target="_self"> Contact Us </a></li><li><a rel="" title="" href="//www.nationalacademies.org/opportunities" target="_self"> Opportunities </a></li><li><a rel="" title="" href="//www.nationalacademies.org/publications" target="_self"> Publications </a></li><li><a rel="" title="" href="//www.nationalacademies.org/about/visiting-us" target="_self"> Visit </a></li><li><a rel="" title="" href="//www.nationalacademies.org/about/operating-status" target="_self"> Current Operating Status </a></li><li><a rel="" title="" href="//www.nationalacademies.org/subscribe" target="_self"> E-Newsletters </a></li></ul> <div class="global-footer__social"><p> Connect With Us </p> <ul><li><a href="https://www.facebook.com/NationalAcademies/" target="_self" class="connect-facebook"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M20 9.375h-2.222c-.334 0-1.111.531-1.111 1.125v2.25H20v3.375h-3.333V24h-3.334v-7.875H10V12.75h3.333V10.5c0-2.482 2.18-4.5 4.515-4.5H20v3.375"></path></svg></a></li><li><a href="https://twitter.com/TheNASEM" target="_self" class="connect-twitter"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M24 9.775a7.238 7.238 0 01-2.121.598 3.787 3.787 0 001.625-2.096 7.253 7.253 0 01-2.347.92A3.65 3.65 0 0018.461 8c-2.04 0-3.695 1.697-3.695 3.787 0 .298.033.586.096.863-3.07-.157-5.793-1.665-7.615-3.956a3.853 3.853 0 00-.499 1.904c0 1.313.853 2.472 1.845 3.151-.606-.02-1.473-.191-1.473-.475v.05c0 1.833 1.072 3.363 2.761 3.712-.31.086-.736.133-1.072.133a3.99 3.99 0 01-.746-.07c.47 1.505 1.81 2.6 3.425 2.632A7.343 7.343 0 016 21.299 10.277 10.277 0 0011.66 23c6.795 0 10.51-5.77 10.51-10.774 0-.163-.005-.328-.013-.491A7.588 7.588 0 0024 9.775"></path></svg></a></li><li><a href="https://www.linkedin.com/company/the-national-academies" target="_self" class="connect-linkedin"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M7 23V10.533h3.4V23H7zm12.467-12.467c2.558 0 4.533 2.034 4.533 5.665V23h-3.4v-6.347c0-1.586-.779-2.679-2.192-2.679-1.082 0-1.94.736-2.223 1.443-.097.251-.118.608-.118.959V23h-3.4s.088-11.333.035-12.467h3.365l-.04 2.267c.516-1.133 1.335-2.267 3.44-2.267zM10.4 7.7a1.7 1.7 0 11-3.4 0 1.7 1.7 0 013.4 0z"></path></svg></a></li><li><a href="http://www.nationalacademies.org/news.xml" target="_self" title="Share our site" class="connect-rss"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path d="M11.85 20.537a2.425 2.425 0 11-4.85 0 2.425 2.425 0 014.85 0zm6.65 1.79c-.316-5.854-5.005-10.548-10.864-10.865a.605.605 0 00-.636.605v1.82c0 .32.246.586.564.607a8.484 8.484 0 017.904 7.904c.02.318.288.564.607.564h1.82c.346 0 .624-.29.605-.636zm5.462.01C23.644 13.484 16.52 6.32 7.625 6A.605.605 0 007 6.606v1.82c0 .327.259.593.585.606 7.238.297 13.048 6.108 13.345 13.345.013.326.28.585.606.585h1.82a.605.605 0 00.606-.625z"></path></svg></a></li><li><a href="//www.nationalacademies.org/subscribe" target="_self" title="Subscribe" class="connect-email"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path fill-rule="evenodd" d="M6 12.964v4.569c0 .296.197.386.439.2l2.822-2.15c.243-.184.243-.484 0-.669l-2.822-2.15c-.242-.185-.439-.095-.439.2zM7.125 9C6.506 9 6 9.601 6 10.071l8.514 6.455a.775.775 0 00.882.002L24 10.07C24 9.601 23.494 9 22.875 9H7.125zm13.613 6.584l2.822 2.15c.243.185.44.095.44-.201v-4.569c0-.295-.197-.385-.44-.2l-2.822 2.15c-.242.185-.242.485 0 .67zM18.26 16.8l-2.822 2.15a.764.764 0 01-.879 0l-2.822-2.15a.765.765 0 00-.88-.001L6 20.495v.29c0 .471.506 1.072 1.125 1.072h15.75c.619 0 1.125-.6 1.125-1.071v-.29L19.141 16.8a.767.767 0 00-.88.001z"></path></svg></a></li></ul></div> <div class="global-footer__legal"><div><button class="button button--icon button--reverse button--secondary"> <a href="https://www8.nationalacademies.org/AcademyGiving/nas/nasform.aspx ">Donate Today</a> <span class="button__icon"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 30 30" class=""><path d="M17.885 10.708a1 1 0 111.412-1.416l5.624 5.608-5.616 5.732a1 1 0 11-1.428-1.4l4.228-4.316-4.22-4.208z"></path><path d="M23.07 14a1 1 0 010 2H7a1 1 0 010-2h16.07z"></path></svg></span></button></div> <div><ul><li><a rel="" title="" href="//www.nationalacademies.org/legal/privacy" target="_self"> Privacy Statement </a></li><li><a rel="" title="" href="//www.nationalacademies.org/legal/policy" target="_self"> DMCA Policy </a></li><li><a rel="" title="" href="//www.nationalacademies.org/legal/terms" target="_self"> Terms of Use </a></li><li><a rel="" title="" href="//www.nationalacademies.org/site-map" target="_self"> Site Map </a></li></ul> <p class="copyright"><a rel="" title="" href="//www.nationalacademies.org/legal" target="_self"> Copyright © <script type="text/javascript"> <!-- var d = new Date(); var n = d.getFullYear(); document.write(n) //--> </script> National Academy of Sciences. All Rights Reserved. </a></p></div></div></div></div></footer> <script src="//nap.nationalacademies.org/includes/js/jquery.dlmenu.js"></script> <script src="//nap.nationalacademies.org/includes/js/jquery.placeholder.js"></script> <!--[if lt IE 9]> <script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script> <![endif]--> <script> // Save to MyNAP handler $('a.saveit').one('click', function(e) { e.preventDefault(); $.getJSON('/napsave.php', { type: 'shelf', r: $(this).data('recordid') }, function(response) { $(e.target) .toggleClass('saveit saveit_active') .attr('href', '/mynap.php') .text('Added to Bookshelf'); }); }); </script> <!-- OneTrust Cookies Consent Notice start for nap.nationalacademies.org --> <script src="https://cdn.cookielaw.org/scripttemplates/otSDKStub.js" type="text/javascript" charset="UTF-8" data-domain-script="5639f7b9-d1de-4e1d-9f2f-e86c3f24727b" ></script> <script type="text/javascript"> function OptanonWrapper() { } </script> <!-- OneTrust Cookies Consent Notice end for nap.nationalacademies.org --> <script src="https://cdnjs.cloudflare.com/ajax/libs/popper.js/1.16.0/umd/popper.min.js"></script> <script src="/includes/js/vendor/bootstrap/4.0.0/bootstrap.bundle.min.js"></script> <script src="/includes/js/bootstrap.min.js"></script> <script> $("#searchagain").click(function() { $(".searchbook").slideToggle("slow", function() { }); }); </script> <!-- 144.171.19.7 --> </body> </html>