CINXE.COM

<!DOCTYPE html><html lang="en-US"><head><meta charSet="utf-8"/><title>Publications - IBM Research</title><meta name="description" content="Browse our catalog of recent publications authored by IBM researchers. This works shows why IBM is one of the most important contributors to modern computing."/><meta name="robots" content="index,follow"/><meta name="viewport" content="width=device-width,initial-scale=1"/><link rel="canonical" href="https://research.ibm.com/publications"/><link rel="icon" href="//www.ibm.com/favicon.ico"/><meta name="dcterms.date" content="2021-02-09"/><meta name="dcterms.rights" content="© Copyright IBM Corp. 2021"/><meta name="geo.country" content="US"/><meta name="google-site-verification" content="O1nsbg1J1iAeYJK6HneffI0_RiLebmSPxfs5ESYNnwI"/><meta property="og:title" content="Publications"/><meta property="og:type" content="website"/><meta property="og:url" content="https://research.ibm.com/publications"/><meta property="og:site_name" content="IBM Research"/><meta property="og:locale" content="en_US"/><meta property="og:description" content="Browse our catalog of recent publications authored by IBM researchers. This works shows why IBM is one of the most important contributors to modern computing."/><meta property="og:image" content="https://research.ibm.com/_next/image?url=https%3A%2F%2Fresearch-website-prod-cms-uploads.s3.us.cloud-object-storage.appdomain.cloud%2Fabout_125b898c36.jpg&w=1200&q=75"/><meta property="og:image:width" content="1200"/><meta property="og:image:height" content="1200"/><meta name="twitter:card" content="summary_large_image"/><meta name="twitter:creator" content="@IBMResearch"/><meta name="twitter:site" content="@IBMResearch"/><meta name="twitter:title" content="Publications"/><meta name="twitter:description" content="Browse our catalog of recent publications authored by IBM researchers. This works shows why IBM is one of the most important contributors to modern computing."/><meta name="twitter:image" content="https://research.ibm.com/_next/image?url=https%3A%2F%2Fresearch-website-prod-cms-uploads.s3.us.cloud-object-storage.appdomain.cloud%2Fabout_125b898c36.jpg&w=1200&q=75"/><meta name="next-head-count" content="26"/><link rel="preload" href="/_next/static/css/69c0d6333f8b5ae9.css" as="style"/><link rel="stylesheet" href="/_next/static/css/69c0d6333f8b5ae9.css" data-n-g=""/><link rel="preload" href="/_next/static/css/dc54a6c9a17471e7.css" as="style"/><link rel="stylesheet" href="/_next/static/css/dc54a6c9a17471e7.css" data-n-p=""/><noscript data-n-css=""></noscript><script defer="" nomodule="" src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js"></script><script src="/_next/static/chunks/webpack-b7fea366180b79b1.js" defer=""></script><script src="/_next/static/chunks/framework-a435b9ecada0512e.js" defer=""></script><script src="/_next/static/chunks/main-9293d76ef0c62f18.js" defer=""></script><script src="/_next/static/chunks/pages/_app-b9582ab63018b5c6.js" defer=""></script><script src="/_next/static/chunks/fc20e1c4-9a05893726f58348.js" defer=""></script><script src="/_next/static/chunks/1c920f04-7653b20eb77a41aa.js" defer=""></script><script src="/_next/static/chunks/5f6bb6fb-d8768b4c13987830.js" defer=""></script><script src="/_next/static/chunks/2f37a77d-c2f4953213066bda.js" defer=""></script><script src="/_next/static/chunks/43a99af2-11b31e1e85eb3ca1.js" defer=""></script><script src="/_next/static/chunks/cfc6b023-9266c81195427c94.js" defer=""></script><script src="/_next/static/chunks/c2ec1d10-a6c9db93ac609649.js" defer=""></script><script src="/_next/static/chunks/e51a3977-d8e768c753798eed.js" defer=""></script><script src="/_next/static/chunks/262b3caa-3d6180fef50adf11.js" defer=""></script><script src="/_next/static/chunks/da011f70-5f6024be6c4e7c77.js" defer=""></script><script src="/_next/static/chunks/42e09a2c-39aeabb415052c30.js" defer=""></script><script src="/_next/static/chunks/761cb607-a1b6fdc8a243c1fe.js" defer=""></script><script src="/_next/static/chunks/2fb713ed-780651db29d294a3.js" defer=""></script><script src="/_next/static/chunks/79eb02f1-42fd67a4ceac70b7.js" defer=""></script><script src="/_next/static/chunks/8f19d780-43022da21c4157b5.js" defer=""></script><script src="/_next/static/chunks/8edc3fe5-3c658903c761796a.js" defer=""></script><script src="/_next/static/chunks/9834-4d5a72b7844c9fa4.js" defer=""></script><script src="/_next/static/chunks/2131-d9b4cb5317eebcae.js" defer=""></script><script src="/_next/static/chunks/2860-6f2f096e239654f6.js" defer=""></script><script src="/_next/static/chunks/3061-1eaaa564aa41acf9.js" defer=""></script><script src="/_next/static/chunks/9755-4a5976c8ec7f097e.js" defer=""></script><script src="/_next/static/chunks/5071-4a74e7be77a5d037.js" defer=""></script><script src="/_next/static/chunks/1527-32861ac0bb15d3aa.js" defer=""></script><script src="/_next/static/chunks/8764-1d5c20476242f5c8.js" defer=""></script><script src="/_next/static/chunks/7177-4dd6fb44e9e6d2b0.js" defer=""></script><script src="/_next/static/chunks/4897-1c982ea37cb2d575.js" defer=""></script><script src="/_next/static/chunks/1235-ae7bb02e277fe668.js" defer=""></script><script src="/_next/static/chunks/1666-d81be63b2c0c107f.js" defer=""></script><script src="/_next/static/chunks/pages/publications-9db69e7e7f922f27.js" defer=""></script><script src="/_next/static/cAHatDzjRLDYxUPqa1puZ/_buildManifest.js" defer=""></script><script src="/_next/static/cAHatDzjRLDYxUPqa1puZ/_ssgManifest.js" defer=""></script></head><body><script>0</script><div id="__next"><div class="bx--masthead "><div class="bx--masthead__l0"><header aria-label="IBM" data-autoid="dds--masthead" class="bx--header"><a class="bx--skip-to-content" href="#main-content" tabindex="0">Skip to main content</a><button data-autoid="dds--masthead-eco-sidenav__l0-menu" aria-label="Open menu" class="bx--masthead__platform bx--header__action bx--header__menu-trigger bx--header__menu-toggle bx--header__menu-toggle__hidden" title="Open menu" type="button"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M2 14.8H18V16H2zM2 11.2H18V12.399999999999999H2zM2 7.6H18V8.799999999999999H2zM2 4H18V5.2H2z"></path></svg></button><div class="bx--side-nav__overlay"></div><nav aria-hidden="true" class="bx--side-nav__navigation bx--side-nav bx--side-nav--ux bx--side-nav--hidden" aria-label="Side navigation"><nav data-autoid="dds--masthead-eco-sidenav__l0"><a data-autoid="dds--masthead-eco-sidenav__l0-productname" href="/" aria-haspopup="true" class="bx--side-nav__submenu bx--side-nav__submenu-platform">Research</a><ul class="bx--side-nav__items"><div class="bx--side-nav__header-navigation"><div><div aria-hidden="false" class="bx--side-nav__menu-section bx--side-nav__menu-section--expanded" id="panel__(-1,-1)"><li class="bx--side-nav__item"><button aria-haspopup="true" aria-expanded="false" aria-selected="" class="bx--side-nav__submenu" data-title="focus-areas" type="button"><div class="bx--side-nav__submenu-content"><span class="bx--side-nav__submenu-title" data-autoid="dds--masthead-eco-sidenav__l0-nav0">Focus areas</span><div class="bx--side-nav__icon bx--side-nav__icon--small bx--side-nav__submenu-chevron"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M16 22L6 12 7.4 10.6 16 19.2 24.6 10.6 26 12z"></path></svg></div></div></button></li><li class="bx--side-nav__menu-item"><a href="/blog" data-autoid="dds--masthead-eco-sidenav__l0-nav1" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Blog</span></a></li><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu--selected"><a href="/publications" data-autoid="dds--masthead-eco-sidenav__l0-nav2" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Publications</span></a></li><li class="bx--side-nav__menu-item"><a href="https://www.ibm.com/careers/us-en/teams/research/" data-autoid="dds--masthead-eco-sidenav__l0-nav3" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Careers</span></a></li><li class="bx--side-nav__item"><button aria-haspopup="true" aria-expanded="false" aria-selected="" class="bx--side-nav__submenu" data-title="about" type="button"><div class="bx--side-nav__submenu-content"><span class="bx--side-nav__submenu-title" data-autoid="dds--masthead-eco-sidenav__l0-nav4">About</span><div class="bx--side-nav__icon bx--side-nav__icon--small bx--side-nav__submenu-chevron"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M16 22L6 12 7.4 10.6 16 19.2 24.6 10.6 26 12z"></path></svg></div></div></button></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(0,-1)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--masthead__side-nav--submemu-title">Focus areas</li><li class="bx--side-nav__menu-item"><a href="/semiconductors" data-autoid="dds--masthead-eco-sidenav__l0-nav0-list0" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Semiconductors</span></a></li><li class="bx--side-nav__menu-item"><a href="/artificial-intelligence" data-autoid="dds--masthead-eco-sidenav__l0-nav0-list1" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Artificial Intelligence</span></a></li><li class="bx--side-nav__menu-item"><a href="/quantum-computing" data-autoid="dds--masthead-eco-sidenav__l0-nav0-list2" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Quantum Computing</span></a></li><li class="bx--side-nav__menu-item"><a href="/hybrid-cloud" data-autoid="dds--masthead-eco-sidenav__l0-nav0-list3" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Hybrid Cloud</span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(4,-1)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--masthead__side-nav--submemu-title">About</li><li class="bx--side-nav__menu-item"><a href="/about" data-autoid="dds--masthead-eco-sidenav__l0-nav4-list0" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Overview</span></a></li><li class="bx--side-nav__menu-item"><a href="/labs" data-autoid="dds--masthead-eco-sidenav__l0-nav4-list1" class="bx--side-nav__link"><span class="bx--side-nav__link-text">Labs</span></a></li><li class="bx--side-nav__menu-item"><a href="/people" data-autoid="dds--masthead-eco-sidenav__l0-nav4-list2" class="bx--side-nav__link"><span class="bx--side-nav__link-text">People</span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(0,0)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/semiconductors" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Semiconductors<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(0,1)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/artificial-intelligence" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Artificial Intelligence<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(0,2)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/quantum-computing" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Quantum Computing<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(0,3)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/hybrid-cloud" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Hybrid Cloud<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(4,0)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/about" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Overview<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(4,1)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/labs" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">Labs<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div><div aria-hidden="true" class="bx--side-nav__menu-section bx--side-nav__menu-section-submenu" id="panel__(4,2)"><button class="bx--masthead__focus" aria-hidden="true"></button><li class="bx--side-nav__menu-item bx--masthead__side-nav--submemu-back"><button class="bx--side-nav__link" data-autoid="dds--masthead-eco-sidenav__l0-back" isbackbutton="true"><span class="bx--side-nav__link-text"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M10 16L20 6 21.4 7.4 12.8 16 21.4 24.6 20 26z"></path></svg>Back</span></button></li><li class="bx--side-nav__item"><a href="/people" class="bx--side-nav__link bx--masthead__side-nav--submemu-section-title"><span class="bx--side-nav__link-text">People<div class="bx--masthead__side-nav--submemu-section-title__icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 20 20" aria-hidden="true"><path d="M11.8 2.8L10.8 3.8 16.2 9.3 1 9.3 1 10.7 16.2 10.7 10.8 16.2 11.8 17.2 19 10z"></path></svg></div></span></a></li><button class="bx--masthead__focus" aria-hidden="true"></button></div></div></div></ul></nav></nav><div class="bx--header__logo"><a aria-label="IBM®" data-autoid="dds--masthead-eco__l0-logo" href="https://www.ibm.com"><img src="/_next/static/media/IBM-8bar-logo--h23.8bbea713.svg" alt="IBM logo"/></a></div><div class="bx--header__search bx--masthead__platform"><div class="bx--masthead__platform-name"><a data-autoid="dds--masthead-eco__l0-ecosystemname" class="bx--header__name" href="/">Research</a></div><div class="bx--header__nav-container"><div class="bx--header__nav-content"><div class="bx--sub-content-left"></div><div class="bx--sub-content-right"></div><nav data-autoid="dds--masthead__l0-nav" aria-label="IBM" class="bx--header__nav"><ul aria-label="IBM" class="bx--header__menu-bar"><li class="bx--header__submenu" data-autoid="dds--masthead-eco__l0-nav0"><a aria-haspopup="menu" aria-expanded="false" class="bx--header__menu-item bx--header__menu-title" data-title="focus-areas" href="#" role="menuitem" tabindex="0" aria-selected="false" aria-label="Focus areas">Focus areas<svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true" class="bx--header__menu-arrow"><path d="M16 22L6 12 7.4 10.6 16 19.2 24.6 10.6 26 12z"></path></svg></a><ul aria-label="Focus areas" class="bx--header__menu"><li><a data-autoid="dds--masthead-eco__l0-nav0--subnav-col0-item0" class="bx--header__menu-item" tabindex="0" href="/semiconductors"><span class="bx--text-truncate--end">Semiconductors</span></a></li><li><a data-autoid="dds--masthead-eco__l0-nav0--subnav-col0-item1" class="bx--header__menu-item" tabindex="0" href="/artificial-intelligence"><span class="bx--text-truncate--end">Artificial Intelligence</span></a></li><li><a data-autoid="dds--masthead-eco__l0-nav0--subnav-col0-item2" class="bx--header__menu-item" tabindex="0" href="/quantum-computing"><span class="bx--text-truncate--end">Quantum Computing</span></a></li><li><a data-autoid="dds--masthead-eco__l0-nav0--subnav-col0-item3" class="bx--header__menu-item" tabindex="0" href="/hybrid-cloud"><span class="bx--text-truncate--end">Hybrid Cloud</span></a></li></ul></li><li><a aria-selected="false" data-autoid="dds--masthead-eco__l0-nav1" class="bx--header__menu-item" tabindex="0" href="/blog"><span class="bx--text-truncate--end">Blog</span></a></li><li><a aria-selected="true" data-autoid="dds--masthead-eco__l0-nav2" class="bx--header__menu-item" tabindex="0" href="/publications"><span class="bx--text-truncate--end">Publications</span></a></li><li><a aria-selected="false" data-autoid="dds--masthead-eco__l0-nav3" class="bx--header__menu-item" tabindex="0" href="https://www.ibm.com/careers/us-en/teams/research/"><span class="bx--text-truncate--end">Careers</span></a></li><li class="bx--header__submenu" data-autoid="dds--masthead-eco__l0-nav4"><a aria-haspopup="menu" aria-expanded="false" class="bx--header__menu-item bx--header__menu-title" data-title="about" href="#" role="menuitem" tabindex="0" aria-selected="false" aria-label="About">About<svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true" class="bx--header__menu-arrow"><path d="M16 22L6 12 7.4 10.6 16 19.2 24.6 10.6 26 12z"></path></svg></a><ul aria-label="About" class="bx--header__menu"><li><a data-autoid="dds--masthead-eco__l0-nav4--subnav-col0-item0" class="bx--header__menu-item" tabindex="0" href="/about"><span class="bx--text-truncate--end">Overview</span></a></li><li><a data-autoid="dds--masthead-eco__l0-nav4--subnav-col0-item1" class="bx--header__menu-item" tabindex="0" href="/labs"><span class="bx--text-truncate--end">Labs</span></a></li><li><a data-autoid="dds--masthead-eco__l0-nav4--subnav-col0-item2" class="bx--header__menu-item" tabindex="0" href="/people"><span class="bx--text-truncate--end">People</span></a></li></ul></li></ul></nav></div><div class="bx--header__nav-caret-left-container" hidden=""><button class="bx--header__nav-caret-left" aria-label="Masthead left caret" tabindex="-1" aria-hidden="true"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M20 24L10 16 20 8z"></path></svg></button><div class="bx--header__nav-caret-left-gradient"></div></div><div class="bx--header__nav-caret-right-container" hidden=""><div class="bx--header__nav-caret-right-gradient"></div><button class="bx--header__nav-caret-right" aria-label="Masthead right caret" tabindex="-1" aria-hidden="true"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M12 8L22 16 12 24z"></path></svg></button></div></div><div class="bx--masthead__overlay"></div><div data-autoid="dds--masthead__search" class="bx--masthead__search"><div class="bx--header__search--actions"><button data-autoid="dds--masthead-eco__l0-search" aria-label="Open IBM search field" tabindex="0" class="bx--header__search--search bx--header__action bx--btn bx--btn--primary bx--btn--icon-only bx--tooltip__trigger bx--tooltip--a11y bx--btn--icon-only--bottom bx--tooltip--align-center" type="button"><div class="bx--assistive-text">Open IBM search field</div><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M29,27.5859l-7.5521-7.5521a11.0177,11.0177,0,1,0-1.4141,1.4141L27.5859,29ZM4,13a9,9,0,1,1,9,9A9.01,9.01,0,0,1,4,13Z"></path></svg></button><button data-autoid="dds--masthead-eco__l0-search--close" aria-label="Close" tabindex="0" class="bx--header__search--close bx--header__action bx--btn bx--btn--primary bx--btn--icon-only bx--tooltip__trigger bx--tooltip--a11y bx--btn--icon-only--bottom bx--tooltip--align-center" type="button"><div class="bx--assistive-text">Close</div><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M24 9.4L22.6 8 16 14.6 9.4 8 8 9.4 14.6 16 8 22.6 9.4 24 16 17.4 22.6 24 24 22.6 17.4 16 24 9.4z"></path></svg></button></div></div></div><div class="bx--header__global bx--header__profile"><button class="bx--header__action bx--overflow-menu" data-autoid="dds--masthead-eco__l0-account" style="width:3rem" type="button" aria-haspopup="true" aria-expanded="false" aria-label="User Profile"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="20" height="20" viewBox="0 0 32 32" aria-hidden="true"><path d="M16 4a5 5 0 11-5 5 5 5 0 015-5m0-2a7 7 0 107 7A7 7 0 0016 2zM26 30H24V25a5 5 0 00-5-5H13a5 5 0 00-5 5v5H6V25a7 7 0 017-7h6a7 7 0 017 7z"></path></svg></button></div></header></div></div><main class="vmZu6" id="main-content"><div class="vbQWi"><div class="Jq8vh oqD_b"><header class="pnjRO"><h1 class="QlIjm">Publications</h1><nav aria-label="breadcrumbs" class="wy3Mq r3uz9"><ol><li><a class="bx--link" href="/">Home</a></li><li><span class="aEWzI" aria-hidden="true">↳ </span><a aria-current="location" href="/publications">Publications</a></li></ol></nav></header></div></div><div class="vbQWi"><div class="ukvEO t_Uew"><div class="N9sLh"><div class="wT9kM"><div role="combobox" aria-haspopup="listbox" aria-owns="react-autowhatever-publications-search" aria-expanded="false" class="vFu9B"><div class="tEbfX"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="_hZsJ"><path d="M29,27.5859l-7.5521-7.5521a11.0177,11.0177,0,1,0-1.4141,1.4141L27.5859,29ZM4,13a9,9,0,1,1,9,9A9.01,9.01,0,0,1,4,13Z"></path></svg><input type="text" autoComplete="off" aria-autocomplete="list" aria-controls="react-autowhatever-publications-search" placeholder="Search titles, abstracts, topics, and authors" role="searchbox" class="t9GPq" value=""/><div class="yR99Q N0vFR"></div><button aria-label="Clear input" class="_2egOn fFsRH bx--btn bx--btn--ghost bx--btn--icon-only" type="button"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" aria-hidden="true" aria-label="Clear input" style="--cds-icon-size-01:1.5rem" width="24" height="24" viewBox="0 0 32 32" role="img" class="bx--btn__icon"><path d="M24 9.4L22.6 8 16 14.6 9.4 8 8 9.4 14.6 16 8 22.6 9.4 24 16 17.4 22.6 24 24 22.6 17.4 16 24 9.4z"></path></svg></button></div><div id="react-autowhatever-publications-search" role="listbox" class="mp_NI"></div></div></div><div class="CoCy7"><div class="_owfA Nnb1a"><div class="aQpJ_"><div class="bx--dropdown__wrapper bx--list-box__wrapper"><label class="bx--label bx--visually-hidden" id="dropdown-toggle-label" for="dropdown-toggle-toggle-button">Filter by</label><div id="dropdown-toggle" class="bx--dropdown bsu3u EcQGl bx--list-box"><button type="button" class="bx--list-box__field" aria-disabled="false" id="dropdown-toggle-toggle-button" aria-haspopup="listbox" aria-expanded="false" aria-labelledby="dropdown-toggle-label dropdown-toggle-toggle-button"><span class="bx--list-box__label">Filter by</span><div class="bx--list-box__menu-icon"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" name="chevron--down" aria-label="Open menu" width="16" height="16" viewBox="0 0 16 16" role="img"><path d="M8 11L3 6 3.7 5.3 8 9.6 12.3 5.3 13 6z"></path><title>Open menu</title></svg></div></button><div id="dropdown-toggle-menu" class="bx--list-box__menu" role="listbox" aria-labelledby="dropdown-toggle-label" tabindex="-1"></div></div></div><div class="yR99Q gzeE0"></div></div></div></div></div></div></div><div class="ZQs4a vbQWi"><div class="sMzQi"><div class="IWzZn"><section aria-labelledby="filter-by-expand-button" id="filter-by" tabindex="-1" class="S_vKZ hv546"><section class="AVelH"><h2 class="_hFje">Date</h2><div class="FnKNf"><div class="DOSa3"><div class="TqqYA"><div class="bx--list-box__wrapper"><label class="bx--label" for="year-range-start" id="year-range-start-label">Start</label><div class="bx--combo-box BNndD bx--list-box"><div class="bx--list-box__field"><input role="combobox" class="bx--text-input bx--text-input--empty" type="text" tabindex="0" aria-autocomplete="list" aria-expanded="false" aria-haspopup="listbox" autoComplete="off" id="year-range-start" placeholder="yyyy" value=""/><button type="button" role="button" aria-label="Open" aria-haspopup="true" data-toggle="true" title="Open" class="bx--list-box__menu-icon" tabindex="-1"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true"><path d="M8 11L3 6 3.7 5.3 8 9.6 12.3 5.3 13 6z"></path></svg></button></div><div id="year-range-start-menu" class="bx--list-box__menu" role="listbox" aria-label="Choose an item"></div></div></div></div></div><div class="jDcOu"><div class="TqqYA"><div class="bx--list-box__wrapper"><label class="bx--label" for="year-range-end" id="year-range-end-label">End</label><div class="bx--combo-box BNndD bx--list-box"><div class="bx--list-box__field"><input role="combobox" class="bx--text-input bx--text-input--empty" type="text" tabindex="0" aria-autocomplete="list" aria-expanded="false" aria-haspopup="listbox" autoComplete="off" id="year-range-end" placeholder="yyyy" value=""/><button type="button" role="button" aria-label="Open" aria-haspopup="true" data-toggle="true" title="Open" class="bx--list-box__menu-icon" tabindex="-1"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true"><path d="M8 11L3 6 3.7 5.3 8 9.6 12.3 5.3 13 6z"></path></svg></button></div><div id="year-range-end-menu" class="bx--list-box__menu" role="listbox" aria-label="Choose an item"></div></div></div></div></div></div></section><section class="AVelH"><h2 class="_hFje">Featured collections</h2><ul class="XBsQU qcmp9 _1PgYZ uvfUZ _0CM3a o0Z_b" style="--gap-sm:1rem;--gap-md:1rem;--gap-lg:0.5rem;--gap-xlg:0.5rem;--gap-max:0.5rem"><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23343"><span title="ICLR 2025">ICLR 2025</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23311"><span title="ICASSP 2025">ICASSP 2025</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23303"><span title="ACS Spring 2025">ACS Spring 2025</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23306"><span title="APS Global Physics Summit 2025">APS Global Physics Summit 2025</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23062"><span title="AAAI 2025">AAAI 2025</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV MmH7c _9ckLP bx--tag--high-contrast bx--tag--interactive" id="filter-sourceInstance-23269"><span title="SPIE Advanced Lithography + Patterning 2025">SPIE Advanced Lithography + Patterning 2025</span></button></li></ul></section><section class="AVelH"><h2 class="_hFje">Popular topics</h2><ul class="XBsQU qcmp9 _1PgYZ uvfUZ _0CM3a o0Z_b" style="--gap-sm:1rem;--gap-md:1rem;--gap-lg:0.5rem;--gap-xlg:0.5rem;--gap-max:0.5rem"><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-ai-hardware"><span title="AI Hardware">AI Hardware</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-foundation-models"><span title="Foundation Models">Foundation Models</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-machine-learning"><span title="Machine Learning">Machine Learning</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-materials-discovery"><span title="Materials Discovery">Materials Discovery</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-quantum-safe-cryptography-and-migration"><span title="Quantum Safe">Quantum Safe</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-quantum-circuits-and-software"><span title="Quantum Software">Quantum Software</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-quantum-hardware"><span title="Quantum Systems">Quantum Systems</span></button></li><li class="iHbnT X4Y7w mBUmN FbtRo _8UAmi bL2oU"><button class="bx--tag ZvHdV Srru3 _9ckLP bx--tag--green bx--tag--interactive" id="filter-tag-semiconductors"><span title="Semiconductors">Semiconductors</span></button></li></ul></section></section></div><div class="UjNFp _3sHhq"><ul><li class="ezC3o" style="--delay:0ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/multivariate-stochastic-dominance-via-optimal-transport-and-applications-to-models-benchmarking">Multivariate Stochastic Dominance via Optimal Transport and Applications to Models Benchmarking</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Gabriel Rioux</li><li class="_8MUnA">Apoorva Nitsure</li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">NeurIPS 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:100ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/tiny-time-mixers-ttms-fast-pre-trained-models-for-enhanced-zerofew-shot-forecasting-of-multivariate-time-series--1">Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA"><a class="WvbPS" href="/people/vijay-e">Vijay E</a></li><li class="_8MUnA"><a class="WvbPS" href="/people/arindam-jati">Arindam Jati</a></li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">NeurIPS 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M31 16L24 23 22.59 21.59 28.17 16 22.59 10.41 24 9 31 16zM1 16L8 9 9.41 10.41 3.83 16 9.41 21.59 8 23 1 16z"></path><path d="M5.91 15H26.080000000000002V17H5.91z" transform="rotate(-75 15.996 16)"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:200ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/distributional-preference-alignment-of-llms-via-optimal-transport">Distributional Preference Alignment of LLMs via Optimal Transport</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Igor Melnyk</li><li class="_8MUnA"><a class="WvbPS" href="/people/youssef-mroueh">Youssef Mroueh</a></li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">NeurIPS 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:300ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/interventional-causal-discovery-in-a-mixture-of-dags">Interventional Causal Discovery in a Mixture of DAGs</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Burak Varici</li><li class="_8MUnA">Dmitriy Katz-Rogozhnikov</li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">NeurIPS 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M31 16L24 23 22.59 21.59 28.17 16 22.59 10.41 24 9 31 16zM1 16L8 9 9.41 10.41 3.83 16 9.41 21.59 8 23 1 16z"></path><path d="M5.91 15H26.080000000000002V17H5.91z" transform="rotate(-75 15.996 16)"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:400ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/graph-transformer-based-surrogate-model-for-accelerated-converter-circuit-topology-design">Graph-Transformer-based Surrogate Model for Accelerated Converter Circuit Topology Design</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Shaoze Fan</li><li class="_8MUnA">Haoshu Lu</li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">DAC 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:500ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/how-do-categorical-duplicates-affect-ml-a-new-benchmark-and-empirical-analyses">How do Categorical Duplicates Affect ML? A New Benchmark and Empirical Analyses</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA"><a class="WvbPS" href="/people/vraj-shah">Vraj Shah</a></li><li class="_8MUnA">Thomas Parashos</li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">VLDB 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:600ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/practical-hamiltonian-monte-carlo-on-riemannian-manifolds-via-relativity-theory">Practical Hamiltonian Monte Carlo on Riemannian Manifolds via Relativity Theory</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Kai Xu</li><li class="_8MUnA">Hong Ge</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">ICML 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:700ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/latent-space-symmetry-discovery--1">Latent Space Symmetry Discovery</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Jianke Yang</li><li class="_8MUnA"><a class="WvbPS" href="/people/nima-dehmamy">Nima Dehmamy</a></li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">ICML 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:800ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/what-would-gauss-say-about-representations-probing-pretrained-image-models-using-synthetic-gaussian-benchmarks">What Would Gauss Say About Representations? Probing Pretrained Image Models using Synthetic Gaussian Benchmarks</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Irene Ko</li><li class="_8MUnA"><a class="WvbPS" href="/people/pin-yu-chen">Pin-Yu Chen</a></li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">ICML 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li><li class="ezC3o" style="--delay:900ms"><article class="mClJ7 PpV0z NybS_ WAhP9"><h2 class="d0naJ _70S6A"><a class="_1kOEn" href="/publications/a-practical-near-optimal-deployment-of-service-function-chains-in-edge-to-cloud-networks">A Practical Near Optimal Deployment of Service Function Chains in Edge-to-Cloud Networks</a></h2><ul class="VgSQD"><li class="EEJK9"><ul class="_8ydsY"><li class="_8MUnA">Rasoul Behravesh</li><li class="_8MUnA"><a class="WvbPS" href="/people/david-breitgand">David Breitgand</a></li><li class="_8MUnA">et al.</li></ul></li><li class="EEJK9">2024</li><li class="EEJK9 XuP8p">INFOCOM 2024</li></ul><div class="FN4MA"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true" class="UI7HX FZFCr"><path d="M17 22L17 14 13 14 13 16 15 16 15 22 12 22 12 24 20 24 20 22 17 22zM16 8a1.5 1.5 0 101.5 1.5A1.5 1.5 0 0016 8z"></path><path d="M16,30A14,14,0,1,1,30,16,14,14,0,0,1,16,30ZM16,4A12,12,0,1,0,28,16,12,12,0,0,0,16,4Z"></path></svg><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 24 24" aria-hidden="true" class="UI7HX QQGoa"><path d="M14 4L12.9 5.1 18.9 11.2 2 11.2 2 12.8 18.9 12.8 12.9 18.9 14 20 22 12z"></path></svg></div></article></li></ul><nav aria-label="pagination" class="_5KtsN mbGgE"><ul class="tgnrU"><li><a aria-current="location" class="QCIlI" href="/publications">1</a></li><li><a class="QCIlI" href="/publications?page=2">2</a></li><li><div class="JGYxh">...</div></li><li><a class="QCIlI" href="/publications?page=300">300</a></li><li><a class="mTtGB" aria-label="Next page" href="/publications?page=2"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="24" height="24" viewBox="0 0 32 32" aria-hidden="true"><path d="M22 16L12 26 10.6 24.6 19.2 16 10.6 7.4 12 6z"></path></svg></a></li></ul></nav></div></div></div></main><footer data-autoid="dds--footer" class="bx--footer"><section class="bx--footer__main"><div class="bx--footer__main-container"><div data-autoid="dds--footer-logo" class="bx--footer-logo"><a data-autoid="dds--footer-logo__link" class="bx--footer-logo__link" href="https://www.ibm.com/"><svg width="157" height="65" class="bx--footer-logo__logo" viewBox="0 0 157 65" xmlns="http://www.w3.org/2000/svg"><title>IBM Logo</title><path d="M30.444 60.208v4.03H0v-4.03h30.444zm78.291-.001v4.03H86.983v-4.03h21.752zm47.858 0v4.03H134.84v-4.03h21.753zm-33.416 0l-1.398 4.03-1.38-4.03h2.778zm-88.384 0h42.775c-2.797 2.426-6.39 3.925-10.327 4.025l-.423.006H34.793v-4.03h42.775zm-4.35-8.46v4.03H0v-4.03h30.444zm52.402 0c-.332 1.248-.8 2.44-1.389 3.555l-.259.474H34.793v-4.029h48.052zm73.748-.005v4.031H134.84v-4.03h21.753zm-47.858 0v4.031H86.983v-4.03h21.752zm17.375 0l-1.398 4.031h-5.85l-1.395-4.03h8.643zM21.745 43.285v4.03H8.698v-4.03h13.047zm61.195 0a17.32 17.32 0 0 1 .476 3.51l.008.52H68.796v-4.03H82.94zm-26.401 0v4.03H43.491v-4.03H56.54zm72.502-.007l-1.396 4.03H115.93l-1.397-4.03h14.507zm18.85 0v4.03h-13.05v-4.03h13.05zm-39.156 0v4.03H95.684v-4.03h13.051zm-86.99-8.454v4.03H8.698v-4.03h13.047zm56.117 0a16.945 16.945 0 0 1 2.926 3.582l.264.447h-37.56v-4.03h34.37zm30.873-.01v4.03H95.684v-4.03h13.051zm39.157 0v4.03H134.84v-4.03h13.052zm-15.919 0l-1.396 4.03h-17.579l-1.396-4.03h20.371zm-50.778-8.452a16.963 16.963 0 0 1-2.82 3.674l-.37.355H43.49v-4.029h37.704zm-59.45 0v4.03H8.698v-4.03h13.047zm126.147-.013v4.031H134.84v-3.839l-1.33 3.839h-11.456l1.373-4.03h24.465zm-27.743 0l1.372 4.031h-11.456l-1.33-3.839v3.84H95.684v-4.032h24.465zm-98.404-8.448v4.03H8.698V17.9h13.047zm61.68 0c0 1.215-.134 2.399-.375 3.542l-.11.487H68.796V17.9h14.628zM56.538 17.9v4.03H43.491V17.9H56.54zm91.352-.015v4.03h-22.954l1.37-4.03h21.584zm-30.624 0l1.372 4.03H95.684v-4.03h21.583zM30.444 9.437v4.03H0v-4.03h30.444zm50.753 0a17.048 17.048 0 0 1 1.498 3.499l.15.531H34.794v-4.03h46.403zm75.396-.018v4.03h-28.776l1.373-4.03h27.403zm-42.207 0l1.372 4.031H86.982V9.42h27.404zM30.444.978v4.03H0V.977h30.444zm36.374 0c3.96 0 7.594 1.415 10.448 3.772l.303.257H34.794V.977h32.024zm89.775-.022v4.031h-25.894l1.372-4.03h24.522zm-45.098 0l1.372 4.03H86.982V.955h24.513z"></path></svg></a></div><nav data-autoid="dds--footer-nav" class="bx--footer-nav"><ul class="bx--accordion bx--footer-nav__container bx--accordion--end"><li class="bx--accordion__item bx--footer-nav-group" data-autoid="dds--footer-nav-group"><button type="button" aria-expanded="false" class="bx--accordion__heading"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true" class="bx--accordion__arrow"><path d="M11 8L6 13 5.3 12.3 9.6 8 5.3 3.7 6 3z"></path></svg><div class="bx--accordion__title">Focus areas</div></button><div class="bx--accordion__content"><h2 class="bx--footer-nav-group__title">Focus areas</h2><ul><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/semiconductors">Semiconductors</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/artificial-intelligence">Artificial Intelligence</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/quantum-computing">Quantum Computing</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/hybrid-cloud">Hybrid Cloud</a></li></ul></div></li><li class="bx--accordion__item bx--footer-nav-group" data-autoid="dds--footer-nav-group"><button type="button" aria-expanded="false" class="bx--accordion__heading"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true" class="bx--accordion__arrow"><path d="M11 8L6 13 5.3 12.3 9.6 8 5.3 3.7 6 3z"></path></svg><div class="bx--accordion__title">Quick links</div></button><div class="bx--accordion__content"><h2 class="bx--footer-nav-group__title">Quick links</h2><ul><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/about">About</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/publications">Publications</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/blog">Blog</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/events">Events</a></li></ul></div></li><li class="bx--accordion__item bx--footer-nav-group" data-autoid="dds--footer-nav-group"><button type="button" aria-expanded="false" class="bx--accordion__heading"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true" class="bx--accordion__arrow"><path d="M11 8L6 13 5.3 12.3 9.6 8 5.3 3.7 6 3z"></path></svg><div class="bx--accordion__title">Work with us</div></button><div class="bx--accordion__content"><h2 class="bx--footer-nav-group__title">Work with us</h2><ul><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="https://www.ibm.com/careers/us-en/teams/research/">Careers</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/contact">Contact Research</a></li></ul></div></li><li class="bx--accordion__item bx--footer-nav-group" data-autoid="dds--footer-nav-group"><button type="button" aria-expanded="false" class="bx--accordion__heading"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true" class="bx--accordion__arrow"><path d="M11 8L6 13 5.3 12.3 9.6 8 5.3 3.7 6 3z"></path></svg><div class="bx--accordion__title">Directories</div></button><div class="bx--accordion__content"><h2 class="bx--footer-nav-group__title">Directories</h2><ul><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/topics">Topics</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/people">People</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="/projects">Projects</a></li></ul></div></li><li class="bx--accordion__item bx--footer-nav-group" data-autoid="dds--footer-nav-group"><button type="button" aria-expanded="false" class="bx--accordion__heading"><svg focusable="false" preserveAspectRatio="xMidYMid meet" xmlns="http://www.w3.org/2000/svg" fill="currentColor" width="16" height="16" viewBox="0 0 16 16" aria-hidden="true" class="bx--accordion__arrow"><path d="M11 8L6 13 5.3 12.3 9.6 8 5.3 3.7 6 3z"></path></svg><div class="bx--accordion__title">Follow us</div></button><div class="bx--accordion__content"><h2 class="bx--footer-nav-group__title">Follow us</h2><ul><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="https://www.ibm.com/account/reg/us-en/signup?formid=news-urx-53237">Newsletter</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="https://twitter.com/ibmresearch">X</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="https://www.linkedin.com/showcase/ibm-research/">LinkedIn</a></li><li class="bx--footer-nav-group__item"><a class="bx--link bx--footer-nav-group__link bx--footer__link" data-autoid="dds--footer-nav-group__link" href="https://www.youtube.com/user/IBMLabs">YouTube</a></li></ul></div></li></ul></nav></div></section><aside data-autoid="dds--footer-legal-nav" class="bx--legal-nav__container"><nav class="bx--legal-nav"><div class="bx--legal-nav__list "><ul class="bx--legal-nav__holder"><li class="bx--legal-nav__list-item"><a href="https://www.ibm.com/contact/us/en/?lnk=flg-cont-usen" class="bx--link bx--footer__link" data-autoid="dds--footer-legal-nav__link-null">Contact IBM</a></li><li class="bx--legal-nav__list-item"><a href="https://www.ibm.com/privacy/us/en/?lnk=flg-priv-usen" class="bx--link bx--footer__link" data-autoid="dds--footer-legal-nav__link-null">Privacy</a></li><li class="bx--legal-nav__list-item"><a href="https://www.ibm.com/us-en/legal?lnk=flg-tous-usen" class="bx--link bx--footer__link" data-autoid="dds--footer-legal-nav__link-null">Terms of use</a></li><li class="bx--legal-nav__list-item"><a href="https://www.ibm.com/accessibility/us/en/?lnk=flg-acce-usen" class="bx--link bx--footer__link" data-autoid="dds--footer-legal-nav__link-null">Accessibility</a></li><li class="bx--legal-nav__list-item" data-autoid="dds--privacy-cp"></li></ul></div></nav></aside></footer></div><script id="__NEXT_DATA__" type="application/json">{"props":{"pageProps":{"initialApolloState":{"UploadFile:866":{"__typename":"UploadFile","alternativeText":"","height":800,"id":"866","url":"https://research-website-prod-cms-uploads.s3.us.cloud-object-storage.appdomain.cloud/about_125b898c36.jpg","width":800},"ComponentSharedSeo:6":{"__typename":"ComponentSharedSeo","canonicalURL":null,"metaDescription":"Browse our catalog of recent publications authored by IBM researchers. This works shows why IBM is one of the most important contributors to modern computing.","id":"6","metaImage":{"__ref":"UploadFile:866"},"metaRobots":null,"metaSocial":[],"metaTitle":"Publications"},"PublicationsPage:1":{"__typename":"PublicationsPage","id":"1","seo":{"__ref":"ComponentSharedSeo:6"}},"ROOT_QUERY":{"__typename":"Query","publicationsPage":{"__ref":"PublicationsPage:1"},"publications({\"after\":\"-1\",\"first\":10,\"sortBy\":\"FEATURED\"})":{"__typename":"QueryPublications_Connection","totalCount":81395,"nodes":[{"__ref":"Publication:155545"},{"__ref":"Publication:154770"},{"__ref":"Publication:155526"},{"__ref":"Publication:155211"},{"__ref":"Publication:153379"},{"__ref":"Publication:154080"},{"__ref":"Publication:153935"},{"__ref":"Publication:150725"},{"__ref":"Publication:151460"},{"__ref":"Publication:117107"}],"featuredFilters":{"__typename":"FeaturedFilters","tags":[{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:8"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:167"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:24"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:78"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:236"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:43"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:41"}},{"__typename":"FeaturedFilterTag","value":{"__ref":"Tag:51"}}],"sourceInstances":[{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23343"}},{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23311"}},{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23303"}},{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23306"}},{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23062"}},{"__typename":"FeaturedFilterSourceInstance","value":{"__ref":"SourceInstance:23269"}}]}}},"Source:453":{"__typename":"Source","id":"453","longName":"Annual Conference on Neural Information Processing Systems","shortName":"NeurIPS","type":"CONFERENCE"},"SourceInstance:22690":{"__typename":"SourceInstance","id":"22690","name":"NeurIPS 2024"},"Author:183986":{"__typename":"Author","id":"183986","ibmer":null},"AuthorName:211596":{"__typename":"AuthorName","id":"211596","firstName":"Gabriel","firstNameInitials":null,"lastName":"Rioux","author":{"__ref":"Author:183986"}},"Affiliation:3020":{"__typename":"Affiliation","id":"3020","isIbm":false},"Author:77386":{"__typename":"Author","id":"77386","ibmer":null},"AuthorName:71556":{"__typename":"AuthorName","id":"71556","firstName":"Apoorva","firstNameInitials":null,"lastName":"Nitsure","author":{"__ref":"Author:77386"}},"Affiliation:73":{"__typename":"Affiliation","id":"73","isIbm":true},"Ibmer:2286":{"__typename":"Ibmer","id":"2286","slug":"mattia-rigotti","displayName":"Mattia Rigotti"},"Author:3672":{"__typename":"Author","id":"3672","ibmer":{"__ref":"Ibmer:2286"}},"AuthorName:4135":{"__typename":"AuthorName","id":"4135","firstName":"Mattia","firstNameInitials":null,"lastName":"Rigotti","author":{"__ref":"Author:3672"}},"Ibmer:1928":{"__typename":"Ibmer","id":"1928","slug":"kristjan-greenewald","displayName":"Kristjan Greenewald"},"Author:18893":{"__typename":"Author","id":"18893","ibmer":{"__ref":"Ibmer:1928"}},"AuthorName:16178":{"__typename":"AuthorName","id":"16178","firstName":"Kristjan","firstNameInitials":null,"lastName":"Greenewald","author":{"__ref":"Author:18893"}},"Ibmer:1754":{"__typename":"Ibmer","id":"1754","slug":"youssef-mroueh","displayName":"Youssef Mroueh"},"Author:18892":{"__typename":"Author","id":"18892","ibmer":{"__ref":"Ibmer:1754"}},"AuthorName:16177":{"__typename":"AuthorName","id":"16177","firstName":"Youssef","firstNameInitials":null,"lastName":"Mroueh","author":{"__ref":"Author:18892"}},"Publication:155545":{"__typename":"Publication","id":"155545","slug":"multivariate-stochastic-dominance-via-optimal-transport-and-applications-to-models-benchmarking","title":"Multivariate Stochastic Dominance via Optimal Transport and Applications to Models Benchmarking","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-12-10","publishedMeta":{"__typename":"PublishedMeta","source":"NeurIPS 2024","year":"2024"},"abstract":"Stochastic dominance is an important concept in probability theory, econometrics and social choice theory for robustly modeling agents' preferences between random outcomes. While many works have been dedicated to the univariate case, little has been done in the multivariate scenario, wherein an agent has to decide between different multivariate outcomes. By exploiting a characterization of multivariate first stochastic dominance in terms of couplings, we introduce a statistic that assesses multivariate almost stochastic dominance under the framework of Optimal Transport with a smooth cost. Further, we introduce an entropic regularization of this statistic, and establish a central limit theorem (CLT) and consistency of the bootstrap procedure for the empirical statistic. Armed with this CLT, we propose a hypothesis testing framework as well as an efficient implementation using the Sinkhorn algorithm. We showcase our method in comparing and benchmarking Large Language Models that are evaluated on multiple metrics. Our multivariate stochastic dominance test allows us to capture the dependencies between the metrics in order to make an informed and statistically significant decision on the relative performance of the models.","linkCode":null,"source":{"__ref":"Source:453"},"sourceInstance":{"__ref":"SourceInstance:22690"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:211596"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:71556"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:4135"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16178"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16177"},"affiliations":[{"__ref":"Affiliation:73"}]}]},"Ibmer:2938":{"__typename":"Ibmer","id":"2938","slug":"vijay-e","displayName":"Vijay E"},"Author:18470":{"__typename":"Author","id":"18470","ibmer":{"__ref":"Ibmer:2938"}},"AuthorName:16104":{"__typename":"AuthorName","id":"16104","firstName":"Vijay","firstNameInitials":null,"lastName":"E","author":{"__ref":"Author:18470"}},"Ibmer:2834":{"__typename":"Ibmer","id":"2834","slug":"arindam-jati","displayName":"Arindam Jati"},"Author:44704":{"__typename":"Author","id":"44704","ibmer":{"__ref":"Ibmer:2834"}},"AuthorName:35793":{"__typename":"AuthorName","id":"35793","firstName":"Arindam","firstNameInitials":null,"lastName":"Jati","author":{"__ref":"Author:44704"}},"Ibmer:500":{"__typename":"Ibmer","id":"500","slug":"pankaj-dayama","displayName":"Pankaj Dayama"},"Author:42731":{"__typename":"Author","id":"42731","ibmer":{"__ref":"Ibmer:500"}},"AuthorName:35110":{"__typename":"AuthorName","id":"35110","firstName":"Pankaj","firstNameInitials":null,"lastName":"Dayama","author":{"__ref":"Author:42731"}},"Author:182633":{"__typename":"Author","id":"182633","ibmer":null},"AuthorName:210243":{"__typename":"AuthorName","id":"210243","firstName":"Sumanta","firstNameInitials":null,"lastName":"Mukherjee","author":{"__ref":"Author:182633"}},"Author:43517":{"__typename":"Author","id":"43517","ibmer":null},"AuthorName:35375":{"__typename":"AuthorName","id":"35375","firstName":"Nam","firstNameInitials":null,"lastName":"Nguyen","author":{"__ref":"Author:43517"}},"Author:44707":{"__typename":"Author","id":"44707","ibmer":null},"AuthorName:35796":{"__typename":"AuthorName","id":"35796","firstName":"Wesley","firstNameInitials":null,"lastName":"Gifford","author":{"__ref":"Author:44707"}},"Author:22962":{"__typename":"Author","id":"22962","ibmer":null},"AuthorName:19187":{"__typename":"AuthorName","id":"19187","firstName":"Chandra","firstNameInitials":null,"lastName":"Reddy","author":{"__ref":"Author:22962"}},"Author:19628":{"__typename":"Author","id":"19628","ibmer":null},"AuthorName:16456":{"__typename":"AuthorName","id":"16456","firstName":"Jayant","firstNameInitials":null,"lastName":"Kalagnanam","author":{"__ref":"Author:19628"}},"Publication:154770":{"__typename":"Publication","id":"154770","slug":"tiny-time-mixers-ttms-fast-pre-trained-models-for-enhanced-zerofew-shot-forecasting-of-multivariate-time-series--1","title":"Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-12-10","publishedMeta":{"__typename":"PublishedMeta","source":"NeurIPS 2024","year":"2024"},"abstract":"Large pre-trained models excel in zero/few-shot learning for language and vision tasks but face challenges in multivariate time series (TS) forecasting due to diverse data characteristics. Consequently, recent research efforts have focused on developing pre-trained TS forecasting models. These models, whether built from scratch or adapted from large language models (LLMs), excel in zero/few-shot forecasting tasks. However, they are limited by slow performance, high computational demands, and neglect of cross-channel and exogenous correlations. To address this, we introduce Tiny Time Mixers (TTM), a compact model (starting from 1M parameters) with effective transfer learning capabilities, trained exclusively on public TS datasets. TTM, based on the light-weight TSMixer architecture, incorporates innovations like adaptive patching, diverse resolution sampling, and resolution prefix tuning to handle pre-training on varied dataset resolutions with minimal model capacity. Additionally, it employs multi-level modeling to capture channel correlations and infuse exogenous signals during fine-tuning. TTM outperforms existing popular benchmarks in zero/few-shot forecasting by (4-40\\%), while reducing computational requirements significantly. Moreover, TTMs are lightweight and can be executed even on CPU-only machines, enhancing usability and fostering wider adoption in resource-constrained environments. The model weights for reproducibility and research use are available at https://huggingface.co/ibm/ttm-research-r2/, while enterprise-use weights under the Apache license can be accessed as follows: the initial $TTM{_Q}$ variant at https://huggingface.co/ibm-granite/granite-timeseries-ttm-r1, and the latest variants $(TTM{_B}, TTM{_E}, TTM{_A})$ weights are available at https://huggingface.co/ibm-granite/granite-timeseries-ttm-r2. The source code for the TTM model along with the usage scripts are available at https://github.com/ibm-granite/granite-tsfm/tree/main/tsfm_public/models/tinytimemixer.","linkCode":"https://huggingface.co/ibm-granite/granite-timeseries-ttm-v1","source":{"__ref":"Source:453"},"sourceInstance":{"__ref":"SourceInstance:22690"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16104"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:35793"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:35110"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:210243"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:35375"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:35796"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:19187"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16456"},"affiliations":[{"__ref":"Affiliation:73"}]}]},"Author:18829":{"__typename":"Author","id":"18829","ibmer":null},"AuthorName:16161":{"__typename":"AuthorName","id":"16161","firstName":"Igor","firstNameInitials":null,"lastName":"Melnyk","author":{"__ref":"Author:18829"}},"Author:33561":{"__typename":"Author","id":"33561","ibmer":null},"AuthorName:27390":{"__typename":"AuthorName","id":"27390","firstName":"Brian","firstNameInitials":null,"lastName":"Belgodere","author":{"__ref":"Author:33561"}},"Ibmer:1522":{"__typename":"Ibmer","id":"1522","slug":"mikhail-yurochkin","displayName":"Mikhail Yurochkin"},"Author:2802":{"__typename":"Author","id":"2802","ibmer":{"__ref":"Ibmer:1522"}},"AuthorName:3526":{"__typename":"AuthorName","id":"3526","firstName":"Mikhail","firstNameInitials":"M.","lastName":"Yurochkin","author":{"__ref":"Author:2802"}},"Author:42288":{"__typename":"Author","id":"42288","ibmer":null},"AuthorName:34946":{"__typename":"AuthorName","id":"34946","firstName":"Jiri","firstNameInitials":null,"lastName":"Navratil","author":{"__ref":"Author:42288"}},"Author:3705":{"__typename":"Author","id":"3705","ibmer":null},"AuthorName:4166":{"__typename":"AuthorName","id":"4166","firstName":"Jerret","firstNameInitials":null,"lastName":"Ross","author":{"__ref":"Author:3705"}},"Publication:155526":{"__typename":"Publication","id":"155526","slug":"distributional-preference-alignment-of-llms-via-optimal-transport","title":"Distributional Preference Alignment of LLMs via Optimal Transport","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-12-10","publishedMeta":{"__typename":"PublishedMeta","source":"NeurIPS 2024","year":"2024"},"abstract":"Current LLM alignment techniques use pairwise human preferences at a sample level, and as such, they do not imply an alignment on the distributional level. We propose in this paper Alignment via Optimal Transport AOT, a novel method for distributional preference alignment of LLMs. AOT aligns LLMs on unpaired preference data by making the reward distribution of the positive samples stochastically dominant in the first order on the distribution of negative samples. We introduce a convex relaxation of this first-order stochastic dominance and cast it as an optimal transport problem with a smooth and convex cost. Thanks to the one-dimensional nature of the resulting optimal transport problem and the convexity of the cost, it has a closed-form solution via sorting on empirical measures. We fine-tune LLMs with this AOT objective, which enables alignment by penalizing the violation of the stochastic dominance of the reward distribution of the positive samples on the reward distribution of the negative samples. We analyze the sample complexity of AOT by considering the dual of the OT problem and show that it converges at the parametric rate. Empirically, we show on a diverse set of alignment datasets and LLMs that AOT leads to state-of-the-art models in the 7B family of models when evaluated with Open LLM Benchmarks and AlpacaEval.","linkCode":null,"source":{"__ref":"Source:453"},"sourceInstance":{"__ref":"SourceInstance:22690"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16161"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16177"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:27390"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:4135"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:71556"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:3526"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16178"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:34946"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:4166"},"affiliations":[{"__ref":"Affiliation:73"}]}]},"Author:19049":{"__typename":"Author","id":"19049","ibmer":null},"AuthorName:16241":{"__typename":"AuthorName","id":"16241","firstName":"Burak","firstNameInitials":null,"lastName":"Varici","author":{"__ref":"Author:19049"}},"Author:42256":{"__typename":"Author","id":"42256","ibmer":null},"AuthorName:34931":{"__typename":"AuthorName","id":"34931","firstName":"Dmitriy","firstNameInitials":null,"lastName":"Katz-Rogozhnikov","author":{"__ref":"Author:42256"}},"Ibmer:819":{"__typename":"Ibmer","id":"819","slug":"dennis-wei","displayName":"Dennis Wei"},"Author:1976":{"__typename":"Author","id":"1976","ibmer":{"__ref":"Ibmer:819"}},"AuthorName:3601":{"__typename":"AuthorName","id":"3601","firstName":"Dennis","firstNameInitials":"D.","lastName":"Wei","author":{"__ref":"Author:1976"}},"Ibmer:4557":{"__typename":"Ibmer","id":"4557","slug":"prasanna-sattigeri","displayName":"Prasanna Sattigeri"},"Author:3590":{"__typename":"Author","id":"3590","ibmer":{"__ref":"Ibmer:4557"}},"AuthorName:11177":{"__typename":"AuthorName","id":"11177","firstName":"Prasanna","firstNameInitials":null,"lastName":"Sattigeri","author":{"__ref":"Author:3590"}},"Author:19269":{"__typename":"Author","id":"19269","ibmer":null},"AuthorName:16332":{"__typename":"AuthorName","id":"16332","firstName":"Ali","firstNameInitials":null,"lastName":"Tajer","author":{"__ref":"Author:19269"}},"Publication:155211":{"__typename":"Publication","id":"155211","slug":"interventional-causal-discovery-in-a-mixture-of-dags","title":"Interventional Causal Discovery in a Mixture of DAGs","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-12-10","publishedMeta":{"__typename":"PublishedMeta","source":"NeurIPS 2024","year":"2024"},"abstract":"Causal interactions among a group of variables are often modeled by a single causal graph. In some domains, however, these interactions are best described by multiple co-existing causal graphs, e.g., in dynamical systems or genomics. This paper addresses the hitherto unknown role of interventions in learning causal interactions among variables governed by a mixture of causal systems, each modeled by one directed acyclic graph (DAG). Causal discovery from mixtures is fundamentally more challenging than single-DAG causal discovery. Two major difficulties stem from (i) inherent uncertainty about the skeletons of the component DAGs that constitute the mixture and (ii) possibly cyclic relationships across these component DAGs. This paper addresses these challenges and aims to identify edges that exist in at least one component DAG of the mixture, referred to as true edges. First, it establishes matching necessary and sufficient conditions on the size of interventions required to identify the true edges. Next, guided by the necessity results, an adaptive algorithm is designed that learns all true edges using O($n^2$) interventions, where n is the number of nodes. Remarkably, the size of the interventions is optimal if the underlying mixture model does not contain cycles across its components. More generally, the gap between the intervention size used by the algorithm and the optimal size is quantified. It is shown to be bounded by the cyclic complexity number of the mixture model, defined as the size of the minimal intervention that can break the cycles in the mixture, which is upper bounded by the number of cycles among the ancestors of a node.","linkCode":"https://github.com/bvarici/intervention-mixture-DAG","source":{"__ref":"Source:453"},"sourceInstance":{"__ref":"SourceInstance:22690"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16241"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:34931"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:3601"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:11177"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16332"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Source:290":{"__typename":"Source","id":"290","longName":"ACM/IEEE Design Automation Conference","shortName":"DAC","type":"CONFERENCE"},"SourceInstance:20707":{"__typename":"SourceInstance","id":"20707","name":"DAC 2024"},"Author:70118":{"__typename":"Author","id":"70118","ibmer":null},"AuthorName:63487":{"__typename":"AuthorName","id":"63487","firstName":"Shaoze","firstNameInitials":"S.","lastName":"Fan","author":{"__ref":"Author:70118"}},"Affiliation:365":{"__typename":"Affiliation","id":"365","isIbm":false},"Author:189429":{"__typename":"Author","id":"189429","ibmer":null},"AuthorName:217595":{"__typename":"AuthorName","id":"217595","firstName":"Haoshu","firstNameInitials":"H.","lastName":"Lu","author":{"__ref":"Author:189429"}},"Author:77150":{"__typename":"Author","id":"77150","ibmer":null},"AuthorName:71296":{"__typename":"AuthorName","id":"71296","firstName":"Shun","firstNameInitials":"S.","lastName":"Zhang","author":{"__ref":"Author:77150"}},"Affiliation:14":{"__typename":"Affiliation","id":"14","isIbm":true},"Author:70121":{"__typename":"Author","id":"70121","ibmer":null},"AuthorName:63490":{"__typename":"AuthorName","id":"63490","firstName":"Ningyuan","firstNameInitials":"N.","lastName":"Cao","author":{"__ref":"Author:70121"}},"Affiliation:117":{"__typename":"Affiliation","id":"117","isIbm":false},"Author:13847":{"__typename":"Author","id":"13847","ibmer":null},"AuthorName:14890":{"__typename":"AuthorName","id":"14890","firstName":"Xin","firstNameInitials":"X.","lastName":"Zhang","author":{"__ref":"Author:13847"}},"Author:1685":{"__typename":"Author","id":"1685","ibmer":null},"AuthorName:1676":{"__typename":"AuthorName","id":"1676","firstName":"Jing","firstNameInitials":"J.","lastName":"Li","author":{"__ref":"Author:1685"}},"Publication:153379":{"__typename":"Publication","id":"153379","slug":"graph-transformer-based-surrogate-model-for-accelerated-converter-circuit-topology-design","title":"Graph-Transformer-based Surrogate Model for Accelerated Converter Circuit Topology Design","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-11-07","publishedMeta":{"__typename":"PublishedMeta","source":"DAC 2024","year":"2024"},"abstract":"Unlike circuit parameter and sizing optimizations, the automated design of analog circuit topologies poses significant challenges for learning-based approaches. One challenge arises from the combinatorial growth of the topology space with circuit size, which limits the topology optimization efficiency. Moreover, traditional circuit evaluation methods are time-consuming, while the presence of data discontinuity in the topology space makes the accurate prediction of circuit performance exceptionally difficult for unseen topologies. To tackle these challenges, we design a novel Graph-Transformer-based Network (GTN) as the surrogate model for circuit evaluation, offering a substantial acceleration in the speed of circuit topology optimization without sacrificing performance. Our GTN model architecture is designed to embed voltage changes in circuit loops and current flows in connected devices, enabling accurate performance predictions for circuits with unseen topologies. Taking the power converter circuit design as an experimental task, our GTN model significantly outperforms an analytical approach and baseline methods directly utilizing graph neural networks. Furthermore, GTN achieves less than 5% relative error and 196× speed-up compared with high-fidelity simulation. Notably, our GTN surrogate model empowers an automatic circuit design framework to discover circuits of comparable quality to those identified through high-fidelity simulation while reducing the time required by up to 98.2%.","linkCode":null,"source":{"__ref":"Source:290"},"sourceInstance":{"__ref":"SourceInstance:20707"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:63487"},"affiliations":[{"__ref":"Affiliation:365"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:217595"},"affiliations":[{"__ref":"Affiliation:365"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:71296"},"affiliations":[{"__ref":"Affiliation:14"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:63490"},"affiliations":[{"__ref":"Affiliation:117"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:14890"},"affiliations":[{"__ref":"Affiliation:14"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:1676"},"affiliations":[{"__ref":"Affiliation:365"}]}]},"Source:473":{"__typename":"Source","id":"473","longName":"International Conference on Very Large Data Bases","shortName":"VLDB","type":"CONFERENCE"},"SourceInstance:22313":{"__typename":"SourceInstance","id":"22313","name":"VLDB 2024"},"Ibmer:5007":{"__typename":"Ibmer","id":"5007","slug":"vraj-shah","displayName":"Vraj Shah"},"Author:181535":{"__typename":"Author","id":"181535","ibmer":{"__ref":"Ibmer:5007"}},"AuthorName:209129":{"__typename":"AuthorName","id":"209129","firstName":"Vraj","firstNameInitials":null,"lastName":"Shah","author":{"__ref":"Author:181535"}},"Author:181536":{"__typename":"Author","id":"181536","ibmer":null},"AuthorName:209130":{"__typename":"AuthorName","id":"209130","firstName":"Thomas","firstNameInitials":null,"lastName":"Parashos","author":{"__ref":"Author:181536"}},"Author:181537":{"__typename":"Author","id":"181537","ibmer":null},"AuthorName:209131":{"__typename":"AuthorName","id":"209131","firstName":"Arun","firstNameInitials":null,"lastName":"Kumar","author":{"__ref":"Author:181537"}},"Publication:154080":{"__typename":"Publication","id":"154080","slug":"how-do-categorical-duplicates-affect-ml-a-new-benchmark-and-empirical-analyses","title":"How do Categorical Duplicates Affect ML? A New Benchmark and Empirical Analyses","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-08-25","publishedMeta":{"__typename":"PublishedMeta","source":"VLDB 2024","year":"2024"},"abstract":"The tedious grunt work involved in data preparation (prep) before ML reduces ML user productivity. It is also a roadblock to industrial-scale cloud AutoML workflows that build ML models for millions of datasets. One important data prep step for ML is cleaning duplicates in the Categorical columns, e.g., deduplicating CA with California in a State column. However, how such Categorical duplicates impact ML is ill-understood as there exist almost no in-depth scientific studies to assess their significance. In this work, we take the first step towards empirically characterizing the impact of Categorical duplicates on ML classification with a three-pronged approach. We first study how Categorical duplicates exhibit themselves by creating a labeled dataset of 1262 Categorical columns. We then curate a downstream benchmark suite of 16 real-world datasets to make observations on the effect of Categorical duplicates on five popular classifiers and five encoding mechanisms. We finally use simulation studies to validate our observations. We find that Logistic Regression and Similarity encoding are more robust to Categorical duplicates than two One-hot encoded high-capacity classifiers. We provide actionable takeaways that can potentially help AutoML developers to build better platforms and ML practitioners to reduce grunt work. While some of the presented insights have remained folklore for practitioners, our work presents the first systematic scientific study to analyze the impact of Categorical duplicates on\nML and put this on an empirically rigorous footing. Our work presents novel data artifacts and benchmarks, as well as novel empirical analyses to spur more research on this topic.","linkCode":null,"source":{"__ref":"Source:473"},"sourceInstance":{"__ref":"SourceInstance:22313"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:209129"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:209130"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:209131"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Source:445":{"__typename":"Source","id":"445","longName":"International Conference on Machine Learning","shortName":"ICML","type":"CONFERENCE"},"SourceInstance:20711":{"__typename":"SourceInstance","id":"20711","name":"ICML 2024"},"Author:181390":{"__typename":"Author","id":"181390","ibmer":null},"AuthorName:208975":{"__typename":"AuthorName","id":"208975","firstName":"Kai","firstNameInitials":null,"lastName":"Xu","author":{"__ref":"Author:181390"}},"Author:181391":{"__typename":"Author","id":"181391","ibmer":null},"AuthorName:208976":{"__typename":"AuthorName","id":"208976","firstName":"Hong","firstNameInitials":null,"lastName":"Ge","author":{"__ref":"Author:181391"}},"Publication:153935":{"__typename":"Publication","id":"153935","slug":"practical-hamiltonian-monte-carlo-on-riemannian-manifolds-via-relativity-theory","title":"Practical Hamiltonian Monte Carlo on Riemannian Manifolds via Relativity Theory","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-07-21","publishedMeta":{"__typename":"PublishedMeta","source":"ICML 2024","year":"2024"},"abstract":"Hamiltonian Monte Carlo (HMC) samples from an unnormalized density by numerically integrating Hamiltonian dynamics. Girolami \u0026 Calderhead (2011) extend HMC to Riemannian manifolds, but the resulting method faces integration instability issues for practical usage. While previous works have tackled this challenge by using more robust metric tensors than Fisher's information metric, our work focuses on designing numerically stable Hamiltonian dynamics. To do so, we start with the idea from Lu et al. (2017), which designs momentum distributions to upper-bound the particle speed. Then, we generalize this Lu et al. (2017) method to Riemannian manifolds. In our generalization, the upper bounds of velocity norm become position-dependent, which intrinsically limits step sizes used in high curvature regions and, therefore, significantly reduces numerical errors. We also derive a more tractable algorithm to sample from relativistic momentum distributions without relying on the mean-field assumption.","linkCode":null,"source":{"__ref":"Source:445"},"sourceInstance":{"__ref":"SourceInstance:20711"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:208975"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:208976"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Author:59375":{"__typename":"Author","id":"59375","ibmer":null},"AuthorName:51601":{"__typename":"AuthorName","id":"51601","firstName":"Jianke","firstNameInitials":null,"lastName":"Yang","author":{"__ref":"Author:59375"}},"Ibmer:537":{"__typename":"Ibmer","id":"537","slug":"nima-dehmamy","displayName":"Nima Dehmamy"},"Author:46331":{"__typename":"Author","id":"46331","ibmer":{"__ref":"Ibmer:537"}},"AuthorName:37422":{"__typename":"AuthorName","id":"37422","firstName":"Nima","firstNameInitials":null,"lastName":"Dehmamy","author":{"__ref":"Author:46331"}},"Author:46332":{"__typename":"Author","id":"46332","ibmer":null},"AuthorName:37423":{"__typename":"AuthorName","id":"37423","firstName":"Robin","firstNameInitials":null,"lastName":"Walters","author":{"__ref":"Author:46332"}},"Author:43403":{"__typename":"Author","id":"43403","ibmer":null},"AuthorName:35328":{"__typename":"AuthorName","id":"35328","firstName":"Rose","firstNameInitials":null,"lastName":"Yu","author":{"__ref":"Author:43403"}},"Publication:150725":{"__typename":"Publication","id":"150725","slug":"latent-space-symmetry-discovery--1","title":"Latent Space Symmetry Discovery","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-07-21","publishedMeta":{"__typename":"PublishedMeta","source":"ICML 2024","year":"2024"},"abstract":"Equivariant neural networks require explicit knowledge of the symmetry group. Automatic symmetry discovery methods aim to relax this constraint and learn invariance and equivariance from data. However, existing symmetry discovery methods are limited to simple linear symmetries and cannot handle the complexity of real-world data. We propose a novel generative model, Latent LieGAN (LaLiGAN), which can discover symmetries of nonlinear group actions. It learns a mapping from the data space to a latent space where the symmetries become linear and simultaneously discovers symmetries in the latent space. Theoretically, we show that our model can express nonlinear symmetries under some conditions about the group action. Experimentally, we demonstrate that our method can accurately discover the intrinsic symmetry in high-dimensional dynamical systems. LaLiGAN also results in a well-structured latent space that is useful for downstream tasks including equation discovery and long-term forecasting.","linkCode":null,"source":{"__ref":"Source:445"},"sourceInstance":{"__ref":"SourceInstance:20711"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:51601"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:37422"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:37423"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:35328"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Author:179971":{"__typename":"Author","id":"179971","ibmer":null},"AuthorName:207208":{"__typename":"AuthorName","id":"207208","firstName":"Irene","firstNameInitials":null,"lastName":"Ko","author":{"__ref":"Author:179971"}},"Ibmer:787":{"__typename":"Ibmer","id":"787","slug":"pin-yu-chen","displayName":"Pin-Yu Chen"},"Author:2143":{"__typename":"Author","id":"2143","ibmer":{"__ref":"Ibmer:787"}},"AuthorName:16022":{"__typename":"AuthorName","id":"16022","firstName":"Pin-Yu","firstNameInitials":"P.Y.","lastName":"Chen","author":{"__ref":"Author:2143"}},"Ibmer:37":{"__typename":"Ibmer","id":"37","slug":"payel-das","displayName":"Payel Das"},"Author:18830":{"__typename":"Author","id":"18830","ibmer":{"__ref":"Ibmer:37"}},"AuthorName:16166":{"__typename":"AuthorName","id":"16166","firstName":"Payel","firstNameInitials":null,"lastName":"Das","author":{"__ref":"Author:18830"}},"Author:46738":{"__typename":"Author","id":"46738","ibmer":null},"AuthorName:37829":{"__typename":"AuthorName","id":"37829","firstName":"Jeet","firstNameInitials":null,"lastName":"Mohapatra","author":{"__ref":"Author:46738"}},"Author:19150":{"__typename":"Author","id":"19150","ibmer":null},"AuthorName:16280":{"__typename":"AuthorName","id":"16280","firstName":"Luca","firstNameInitials":null,"lastName":"Daniel","author":{"__ref":"Author:19150"}},"Publication:151460":{"__typename":"Publication","id":"151460","slug":"what-would-gauss-say-about-representations-probing-pretrained-image-models-using-synthetic-gaussian-benchmarks","title":"What Would Gauss Say About Representations? Probing Pretrained Image Models using Synthetic Gaussian Benchmarks","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-07-21","publishedMeta":{"__typename":"PublishedMeta","source":"ICML 2024","year":"2024"},"abstract":"Recent years have witnessed a paradigm shift in deep learning from task-centric model design to task-agnostic representation learning and task-specific fine-tuning. Pretrained model representations are commonly evaluated extensively across various real-world tasks and used as a foundation for different downstream tasks. This paper proposes a solution for assessing the quality of representations in a task-agnostic way. To circumvent the need for real-world data in evaluation, we explore the use of synthetic binary classification tasks with Gaussian mixtures to probe pretrained models and compare the robustness-accuracy performance on pretrained representations with an idealized reference. Our approach offers a holistic evaluation, revealing intrinsic model capabilities and reducing the dependency on real-life data for model evaluation. Evaluated with various pretrained image models, the experimental results confirm that our task-agnostic evaluation correlates with actual linear probing performance on downstream tasks and can also guide parameter choice in robust linear probing to achieve a better robustness-accuracy trade-off.","linkCode":null,"source":{"__ref":"Source:445"},"sourceInstance":{"__ref":"SourceInstance:20711"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:207208"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16022"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16166"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:37829"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16280"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Source:299":{"__typename":"Source","id":"299","longName":"IEEE Conference on Computer Communications","shortName":"INFOCOM","type":"CONFERENCE"},"SourceInstance:12142":{"__typename":"SourceInstance","id":"12142","name":"INFOCOM 2024"},"Author:156795":{"__typename":"Author","id":"156795","ibmer":null},"AuthorName:174443":{"__typename":"AuthorName","id":"174443","firstName":"Rasoul","firstNameInitials":null,"lastName":"Behravesh","author":{"__ref":"Author:156795"}},"Ibmer:1542":{"__typename":"Ibmer","id":"1542","slug":"david-breitgand","displayName":"David Breitgand"},"Author:58225":{"__typename":"Author","id":"58225","ibmer":{"__ref":"Ibmer:1542"}},"AuthorName:50451":{"__typename":"AuthorName","id":"50451","firstName":"David","firstNameInitials":null,"lastName":"Breitgand","author":{"__ref":"Author:58225"}},"Ibmer:1131":{"__typename":"Ibmer","id":"1131","slug":"dean-lorenz","displayName":"Dean Lorenz"},"Author:20387":{"__typename":"Author","id":"20387","ibmer":{"__ref":"Ibmer:1131"}},"AuthorName:16695":{"__typename":"AuthorName","id":"16695","firstName":"Dean","firstNameInitials":null,"lastName":"Lorenz","author":{"__ref":"Author:20387"}},"Author:156796":{"__typename":"Author","id":"156796","ibmer":null},"AuthorName:174444":{"__typename":"AuthorName","id":"174444","firstName":"Danny","firstNameInitials":null,"lastName":"Raz","author":{"__ref":"Author:156796"}},"Publication:117107":{"__typename":"Publication","id":"117107","slug":"a-practical-near-optimal-deployment-of-service-function-chains-in-edge-to-cloud-networks","title":"A Practical Near Optimal Deployment of Service Function Chains in Edge-to-Cloud Networks","type":{"__typename":"PublicationType","displayValue":"Conference paper"},"published":"2024-05-20","publishedMeta":{"__typename":"PublishedMeta","source":"INFOCOM 2024","year":"2024"},"abstract":"Mobile edge computing offers a myriad of opportunities to innovate and introduce novel applications, thereby enhancing user experiences considerably. A critical issue extensively investigated in this domain is efficient deployment of Service Function Chains (SFCs) across the physical network, spanning from the edge to the cloud. This problem is known to be NP-hard. As a result of its practical importance, there is significant interest in the development of high-quality sub-optimal solutions. In this paper, we consider this problem and propose a novel near-optimal heuristic that is extremely efficient and scalable.\nWe compare our solution to the state-of-the-art heuristics and to the theoretical optimum. In our large scale evaluations, we use\nrealistic topologies which are previously reported in the literature. We demonstrate that the execution time offered by our solution\ngrows slowly as the number of Virtual Network Function (VNF) forwarding graph embedding requests grows, and it handles one million requests in slightly more than 20 seconds for 100 nodes and 150 edges physical topology.","linkCode":null,"source":{"__ref":"Source:299"},"sourceInstance":{"__ref":"SourceInstance:12142"},"authors":[{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:174443"},"affiliations":[{"__ref":"Affiliation:3020"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:50451"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:16695"},"affiliations":[{"__ref":"Affiliation:73"}]},{"__typename":"PublicationHasAuthorName","authorName":{"__ref":"AuthorName:174444"},"affiliations":[{"__ref":"Affiliation:3020"}]}]},"Tag:8":{"__typename":"Tag","id":"8","name":"AI Hardware","slug":"ai-hardware"},"Tag:167":{"__typename":"Tag","id":"167","name":"Foundation Models","slug":"foundation-models"},"Tag:24":{"__typename":"Tag","id":"24","name":"Machine Learning","slug":"machine-learning"},"Tag:78":{"__typename":"Tag","id":"78","name":"Materials Discovery","slug":"materials-discovery"},"Tag:236":{"__typename":"Tag","id":"236","name":"Quantum Safe","slug":"quantum-safe-cryptography-and-migration"},"Tag:43":{"__typename":"Tag","id":"43","name":"Quantum Software","slug":"quantum-circuits-and-software"},"Tag:41":{"__typename":"Tag","id":"41","name":"Quantum Systems","slug":"quantum-hardware"},"Tag:51":{"__typename":"Tag","id":"51","name":"Semiconductors","slug":"semiconductors"},"SourceInstance:23343":{"__typename":"SourceInstance","id":"23343","name":"ICLR 2025"},"SourceInstance:23311":{"__typename":"SourceInstance","id":"23311","name":"ICASSP 2025"},"SourceInstance:23303":{"__typename":"SourceInstance","id":"23303","name":"ACS Spring 2025"},"SourceInstance:23306":{"__typename":"SourceInstance","id":"23306","name":"APS Global Physics Summit 2025"},"SourceInstance:23062":{"__typename":"SourceInstance","id":"23062","name":"AAAI 2025"},"SourceInstance:23269":{"__typename":"SourceInstance","id":"23269","name":"SPIE Advanced Lithography + Patterning 2025"}}},"__N_SSP":true},"page":"/publications","query":{},"buildId":"cAHatDzjRLDYxUPqa1puZ","isFallback":false,"gssp":true,"locale":"en-US","locales":["en-US"],"defaultLocale":"en-US","scriptLoader":[]}</script></body></html>