CINXE.COM

<!DOCTYPE html><html lang="en"> <head itemscope itemtype="https://schema.org/WebSite"> <meta charset="utf-8"> <meta http-equiv="X-UA-Compatible" content="IE=edge"> <meta name="viewport" content="width=device-width, initial-scale=1"> <base href="/"/> <title>Deep Learning | Bosch Center for Artificial Intelligence</title> <meta name="description" content="Our Fields Of Expertise - Deep Learning "> <meta name="DCSext.wtg_lang" content="en" /> <meta itemprop="name" content="Bosch Center for Artificial Intelligence"/> <meta itemprop="url" content="https://www.bosch-ai.com/"/>  <meta name="theme-color" content="#ffffff">  <link rel="shortcut icon" type="image/x-icon" href="/media/tech/tech_images/favicon.ico"> <link rel="icon" type="image/png" href="/media/tech/tech_images/favicon-png.png" sizes="16x16">  <link rel="canonical" href="https://www.bosch-ai.com/research/fields-of-expertise/deep-learning/"> <link rel="alternate" href="https://www.bosch-ai.com/research/fields-of-expertise/deep-learning/" hreflang="en-EN" /> <link itemprop="url" href="https://www.bosch-ai.com/research/fields-of-expertise/deep-learning/"/>  <link rel="preload" href="/media/tech/fonts/woff2/boschsans_regular.woff2" as="font" type="font/woff2" crossorigin> <link rel="preload" href="/media/tech/fonts/woff2/boschsans_bold.woff2" as="font" type="font/woff2" crossorigin>  <link rel="stylesheet" href="/media/tech/styles/main.css?ver=1716558076529">  <meta property="og:title" content="Deep Learning" /> <meta property="og:site_name" content="Bosch Center for Artificial Intelligence" /> <meta property="og:url" content="https://www.bosch-ai.com/research/fields-of-expertise/deep-learning/" /> <meta property="og:description" content="Our Fields Of Expertise - Deep Learning" /> <meta property="og:image" content="https://www.bosch-ai.com/media/reserach/research_fields/202102_ai_renningen_automated_driving_layer_1.jpg" /> <meta property="og:type" content="website" /> <meta property="fb:app_id" content="null" />  <meta name="twitter:card" content="summary_large_image"> <meta name="twitter:site" content="@Bosch"> <meta name="twitter:creator" content="@Bosch"> <meta name="twitter:title" content="Deep Learning"> <meta name="twitter:description" content="Our Fields Of Expertise - Deep Learning"> <meta name="twitter:image" content="https://www.bosch-ai.com/media/reserach/research_fields/202102_ai_renningen_automated_driving_layer_1.jpg">  <meta name="msapplication-tap-highlight" content="no">  <meta name="msapplication-TileImage" content=""> <meta name="msapplication-TileColor" content="#ffffff"> <script type="application/json" id="js-externalConfig"> { "defaultLocalization": { "defaultLanguage": "en", "defaultCountry": "UK", "defaultCaaSLanguage": "en", "defaultDateFormat": "MM/dd/yyyy" } , "search": { "baseUrl": "https://search.internet.bosch.com/RB.SEARCH/search/bosch-ai/search.json?doc_lang=en&lang=de&q=ALL&scenario=1", "targetPath": "/search.html" }, "socialMediaShare": { "facebookUrl": "https://www.facebook.com/dialog/feed", "twitterUrl": "https://www.twitter.com/intent/tweet", "linkedinUrl": "https://www.linkedin.com/shareArticle" }, "schemaTeasers": { "author": "Bosch Global", "publisher": "Bosch Global", "logoUrl": "/media/tech/tech_images/bosch_logo_res_260x85.webp" } } </script> </head> <body itemscope itemtype="https://schema.org/Organization"> <div id="fb-root"></div> <header class="O-Header -primary">  <a href="#main" class="a-link--skip-to-main-content">Skip to main content</a>  <div class="O-Header__wrapper" itemscope itemtype="https://schema.org/Brand" itemprop="brand"> <nav class="O-Header__mainRow e-container -primary" aria-label="Quicklinks"> <a itemprop="url" class="O-Header__brandLogo" target="_self" aria-label="Home" href="/"><svg xmlns="http://www.w3.org/2000/svg" width="433" height="97" viewBox="0 0 433 97"><g><g><path d="M185.2,46.88a13.77,13.77,0,0,0,8.8-13c0-11.7-8.3-17.5-19.7-17.5H144.4V80h32.5c10,0,19.8-7,19.8-17.7C196.7,49.58,185.2,47,185.2,46.88ZM160,29.58h11.6a5.66,5.66,0,0,1,6,5.31q0,.34,0,.69a5.93,5.93,0,0,1-6,5.81H159.9Zm11.7,37.1H160.1V54.18h11.3c5.7,0,8.4,2.5,8.4,6.2C179.8,65,176.4,66.68,171.7,66.68Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M231.1,14.78c-18.4,0-29.2,14.7-29.2,33.3s10.8,33.3,29.2,33.3,29.2-14.6,29.2-33.3S249.6,14.78,231.1,14.78Zm0,51.4c-9,0-13.5-8.1-13.5-18.1s4.5-18,13.5-18,13.6,8.1,13.6,18C244.7,58.18,240.1,66.18,231.1,66.18Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M294.2,41.38l-2.2-.5c-5.4-1.1-9.7-2.5-9.7-6.4,0-4.2,4.1-5.9,7.7-5.9a17.86,17.86,0,0,1,13,5.9l9.9-9.8c-4.5-5.1-11.8-10-23.2-10-13.4,0-23.6,7.5-23.6,20,0,11.4,8.2,17,18.2,19.1l2.2.5c8.3,1.7,11.4,3,11.4,7,0,3.8-3.4,6.3-8.6,6.3-6.2,0-11.8-2.7-16.1-8.2l-10.1,10c5.6,6.7,12.7,11.9,26.4,11.9,11.9,0,24.6-6.8,24.6-20.7C314.3,46.08,303.3,43.28,294.2,41.38Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M349.7,66.18c-7,0-14.3-5.8-14.3-18.5,0-11.3,6.8-17.6,13.9-17.6,5.6,0,8.9,2.6,11.5,7.1l12.8-8.5c-6.4-9.7-14-13.8-24.5-13.8-19.2,0-29.6,14.9-29.6,32.9,0,18.9,11.5,33.7,29.4,33.7,12.6,0,18.6-4.4,25.1-13.8L361.1,59C358.5,63.18,355.7,66.18,349.7,66.18Z" style="fill:#ed0007;fill-rule:evenodd"></path><polygon points="416.3 16.38 416.3 39.78 397 39.78 397 16.38 380.3 16.38 380.3 79.98 397 79.98 397 54.88 416.3 54.88 416.3 79.98 433 79.98 433 16.38 416.3 16.38" style="fill:#ed0007;fill-rule:evenodd"></polygon></g><g><path d="M48.2.18a48.2,48.2,0,1,0,48.2,48.2A48.2,48.2,0,0,0,48.2.18Zm0,91.9a43.7,43.7,0,1,1,43.7-43.7,43.71,43.71,0,0,1-43.7,43.7Z"></path><path d="M68.1,18.28H64.8v16.5H31.7V18.28H28.3a36.06,36.06,0,0,0,0,60.2h3.4V62H64.8v16.5h3.3a36.05,36.05,0,0,0,0-60.2ZM27.1,72A31.59,31.59,0,0,1,24.47,27.4a32.51,32.51,0,0,1,2.63-2.62Zm37.7-14.6H31.7V39.28H64.8Zm4.5,14.5v-10h0V34.78h0v-10a31.65,31.65,0,0,1,2.39,44.71A33.68,33.68,0,0,1,69.3,71.88Z"></path></g></g></svg></a> <div class="O-Header__mainRow__item O-Header__mainRow__toggle"> <div class="M-Main-Navigation" data="true" data-js-component="main-navigation"> <div class="M-Main-Navigation__buttons"> <div class="M-Main-Navigation__buttons--link"> <a class="a-button--integrated" target="_self" aria-label="Contact us" href="/contact/"><i class="a-icon boschicon-bosch-ic-chat" title="Contact us"></i></a> </div> <button type="burger" class="a-button a-button--integrated -without-label M-Main-Navigation__buttons--toggle-search" aria-label="Search" > <i class="a-icon boschicon-bosch-ic-search a-button__icon" title="Search"></i> </button> <button type="button" class="M-Main-Navigation__toggle -primary" title="Menu" aria-label="Open Main Navigation" aria-haspopup="true"> <span class="M-Main-Navigation__toggle__icon"> <span class="M-Main-Navigation__toggle__bar"></span> <span class="M-Main-Navigation__toggle__bar"></span> <span class="M-Main-Navigation__toggle__bar"></span> <span class="M-Main-Navigation__toggle__bar"></span> </span> </button> </div> <nav class="M-Main-Navigation__flyoutWrapper -primary" aria-label="Main Navigation"> <div class="M-Main-Navigation__flyoutWrapper--scroll-container"> <div class="M-Main-Navigation__flyoutWrapper--inner"> <ul class="M-Main-Navigation__items M-Main-Navigation__items--first" role="menubar"> <li class="M-Main-Navigation__item M-Main-Navigation__item--first" data-is-active="true" role="none" data-id="research"> <a class="M-Main-Navigation__link M-Main-Navigation__linkToSecond" aria-haspopup="true" role="menuitem" tabindex="-1"><span> Research </span> <i class="a-icon ui-ic-forward" title="foward"></i></a> <ul class="M-Main-Navigation__items M-Main-Navigation__items--second" aria-hidden="true" aria-expanded="false" role="menu"> <li class="M-Main-Navigation__item M-Main-Navigation__backLink" role="menuitem"> <button type="burger" class="a-button a-button a-button--integrated" aria-label="Menu back" > <i class="a-icon ui-ic-backward a-button__icon" title="backward"></i> <span class="a-button__label"> Main Navigation </span> </button> </li> <li class="M-Main-Navigation__item" role="none"> <span class="M-Main-Navigation__link--title" role="menuitem" tabindex="-1"> Research </span> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" data-is-active="true" role="none"> <a class="M-Main-Navigation__link" href="/research/fields-of-expertise/" role="menuitem" tabindex="-1"><span>Fields Of Expertise </span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/research/collaborations/" role="menuitem" tabindex="-1"><span>Collaborations</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/research/publications/" role="menuitem" tabindex="-1"><span>Publications</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/research/scientific-advisory-board/" role="menuitem" tabindex="-1"><span>Scientific Advisory Board</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/research/publicly-funded-projects/" role="menuitem" tabindex="-1"><span>Publicly Funded Projects</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/research/distinguished-lecture-series/" role="menuitem" tabindex="-1"><span>Distinguished Lecture Series</span></a> </li> </ul> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--first" role="none" data-id="ai_applied"> <a class="M-Main-Navigation__link M-Main-Navigation__linkToSecond" aria-haspopup="true" role="menuitem"><span> Industrial AI </span> <i class="a-icon ui-ic-forward" title="foward"></i></a> <ul class="M-Main-Navigation__items M-Main-Navigation__items--second" aria-hidden="true" aria-expanded="false" role="menu"> <li class="M-Main-Navigation__item M-Main-Navigation__backLink" role="menuitem"> <button type="burger" class="a-button a-button a-button--integrated" aria-label="Menu back" > <i class="a-icon ui-ic-backward a-button__icon" title="backward"></i> <span class="a-button__label"> Main Navigation </span> </button> </li> <li class="M-Main-Navigation__item" role="none"> <span class="M-Main-Navigation__link--title" role="menuitem" tabindex="-1"> Industrial AI </span> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/industrial-ai/" role="menuitem" tabindex="-1"> <span>Overview</span> </a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/industrial-ai/code-of-ethics-for-ai/" role="menuitem" tabindex="-1"><span>Code of Ethics for AI </span></a> </li> </ul> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--first" role="none" data-id="events"> <a class="M-Main-Navigation__link M-Main-Navigation__linkToSecond" aria-haspopup="true" role="menuitem"><span> Bosch AI CON </span> <i class="a-icon ui-ic-forward" title="foward"></i></a> <ul class="M-Main-Navigation__items M-Main-Navigation__items--second" aria-hidden="true" aria-expanded="false" role="menu"> <li class="M-Main-Navigation__item M-Main-Navigation__backLink" role="menuitem"> <button type="burger" class="a-button a-button a-button--integrated" aria-label="Menu back" > <i class="a-icon ui-ic-backward a-button__icon" title="backward"></i> <span class="a-button__label"> Main Navigation </span> </button> </li> <li class="M-Main-Navigation__item" role="none"> <span class="M-Main-Navigation__link--title" role="menuitem" tabindex="-1"> Bosch AI CON </span> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/bosch-ai-con/ai-con/" role="menuitem" tabindex="-1"><span>AI CON </span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/bosch-ai-con/bosch-ai-con-2018/" role="menuitem" tabindex="-1"><span>Bosch AI CON 2018</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/bosch-ai-con/bosch-ai-con-2019/" role="menuitem" tabindex="-1"><span>Bosch AI CON 2019</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/bosch-ai-con/bosch-ai-con-2021/" role="menuitem" tabindex="-1"><span>Bosch AI CON 2021</span></a> </li> </ul> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--first" role="none" data-id="about_us"> <a class="M-Main-Navigation__link M-Main-Navigation__linkToSecond" aria-haspopup="true" role="menuitem"><span> About Us </span> <i class="a-icon ui-ic-forward" title="foward"></i></a> <ul class="M-Main-Navigation__items M-Main-Navigation__items--second" aria-hidden="true" aria-expanded="false" role="menu"> <li class="M-Main-Navigation__item M-Main-Navigation__backLink" role="menuitem"> <button type="burger" class="a-button a-button a-button--integrated" aria-label="Menu back" > <i class="a-icon ui-ic-backward a-button__icon" title="backward"></i> <span class="a-button__label"> Main Navigation </span> </button> </li> <li class="M-Main-Navigation__item" role="none"> <span class="M-Main-Navigation__link--title" role="menuitem" tabindex="-1"> About Us </span> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/about-us/about-us/" role="menuitem" tabindex="-1"><span>About Us</span></a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/about-us/our-locations/" role="menuitem" tabindex="-1"><span>Our Locations</span></a> </li> </ul> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--first" role="none" data-id="we_are_hiring"> <a class="M-Main-Navigation__link M-Main-Navigation__linkToSecond" aria-haspopup="true" role="menuitem"><span> Careers </span> <i class="a-icon ui-ic-forward" title="foward"></i></a> <ul class="M-Main-Navigation__items M-Main-Navigation__items--second" aria-hidden="true" aria-expanded="false" role="menu"> <li class="M-Main-Navigation__item M-Main-Navigation__backLink" role="menuitem"> <button type="burger" class="a-button a-button a-button--integrated" aria-label="Menu back" > <i class="a-icon ui-ic-backward a-button__icon" title="backward"></i> <span class="a-button__label"> Main Navigation </span> </button> </li> <li class="M-Main-Navigation__item" role="none"> <span class="M-Main-Navigation__link--title" role="menuitem" tabindex="-1"> Careers </span> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/careers/" role="menuitem" tabindex="-1"> <span>Overview</span> </a> </li> <li class="M-Main-Navigation__item M-Main-Navigation__item--second" role="none"> <a class="M-Main-Navigation__link" href="/careers/promotion-of-young-talents/" role="menuitem" tabindex="-1"><span>Promotion of Young Talents</span></a> </li> </ul> </li> </ul> </div> </div> </nav> <div id="M-Main-Navigation__search-flyout" class="M-Search-Flyout"> <div class="M-Search-Flyout__wrapper"> <div class="e-container"> <div class="M-Search-Flyout__header-wrapper"> <a itemprop="url" class="M-Search-Flyout__header-brand-logo" target="_self" aria-label="Home" href="/"><svg xmlns="http://www.w3.org/2000/svg" width="433" height="97" viewBox="0 0 433 97"><g><g><path d="M185.2,46.88a13.77,13.77,0,0,0,8.8-13c0-11.7-8.3-17.5-19.7-17.5H144.4V80h32.5c10,0,19.8-7,19.8-17.7C196.7,49.58,185.2,47,185.2,46.88ZM160,29.58h11.6a5.66,5.66,0,0,1,6,5.31q0,.34,0,.69a5.93,5.93,0,0,1-6,5.81H159.9Zm11.7,37.1H160.1V54.18h11.3c5.7,0,8.4,2.5,8.4,6.2C179.8,65,176.4,66.68,171.7,66.68Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M231.1,14.78c-18.4,0-29.2,14.7-29.2,33.3s10.8,33.3,29.2,33.3,29.2-14.6,29.2-33.3S249.6,14.78,231.1,14.78Zm0,51.4c-9,0-13.5-8.1-13.5-18.1s4.5-18,13.5-18,13.6,8.1,13.6,18C244.7,58.18,240.1,66.18,231.1,66.18Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M294.2,41.38l-2.2-.5c-5.4-1.1-9.7-2.5-9.7-6.4,0-4.2,4.1-5.9,7.7-5.9a17.86,17.86,0,0,1,13,5.9l9.9-9.8c-4.5-5.1-11.8-10-23.2-10-13.4,0-23.6,7.5-23.6,20,0,11.4,8.2,17,18.2,19.1l2.2.5c8.3,1.7,11.4,3,11.4,7,0,3.8-3.4,6.3-8.6,6.3-6.2,0-11.8-2.7-16.1-8.2l-10.1,10c5.6,6.7,12.7,11.9,26.4,11.9,11.9,0,24.6-6.8,24.6-20.7C314.3,46.08,303.3,43.28,294.2,41.38Z" style="fill:#ed0007;fill-rule:evenodd"></path><path d="M349.7,66.18c-7,0-14.3-5.8-14.3-18.5,0-11.3,6.8-17.6,13.9-17.6,5.6,0,8.9,2.6,11.5,7.1l12.8-8.5c-6.4-9.7-14-13.8-24.5-13.8-19.2,0-29.6,14.9-29.6,32.9,0,18.9,11.5,33.7,29.4,33.7,12.6,0,18.6-4.4,25.1-13.8L361.1,59C358.5,63.18,355.7,66.18,349.7,66.18Z" style="fill:#ed0007;fill-rule:evenodd"></path><polygon points="416.3 16.38 416.3 39.78 397 39.78 397 16.38 380.3 16.38 380.3 79.98 397 79.98 397 54.88 416.3 54.88 416.3 79.98 433 79.98 433 16.38 416.3 16.38" style="fill:#ed0007;fill-rule:evenodd"></polygon></g><g><path d="M48.2.18a48.2,48.2,0,1,0,48.2,48.2A48.2,48.2,0,0,0,48.2.18Zm0,91.9a43.7,43.7,0,1,1,43.7-43.7,43.71,43.71,0,0,1-43.7,43.7Z"></path><path d="M68.1,18.28H64.8v16.5H31.7V18.28H28.3a36.06,36.06,0,0,0,0,60.2h3.4V62H64.8v16.5h3.3a36.05,36.05,0,0,0,0-60.2ZM27.1,72A31.59,31.59,0,0,1,24.47,27.4a32.51,32.51,0,0,1,2.63-2.62Zm37.7-14.6H31.7V39.28H64.8Zm4.5,14.5v-10h0V34.78h0v-10a31.65,31.65,0,0,1,2.39,44.71A33.68,33.68,0,0,1,69.3,71.88Z"></path></g></g></svg></a> <div class="M-Search-Flyout__button-wrapper"> <button type="burger" class="a-button a-button a-button--integrated M-Search-Flyout--close-button" aria-label="Close" id="flyout-wrapper-full" > <i class="a-icon boschicon-bosch-ic-close a-button__icon" title="Close"></i> <span class="a-button__label"> Close </span> </button> </div> </div> </div> <div data-js-component="SearchResultGroup" data-land-mark="flyout" data-flyout-mode="true" data-top-query-headline="" data-placeholder="Search" data-aria-label-search-icon="Search" data-aria-label-search-form="Search header" data-aria-label-close-icon="Close" data-autosuggestions-link="See all results" data-hide-number-of-results="true" data-search-url="/search.html" > </div> </div> </div> </div> </div> </nav> </div> <div class="O-Header__navRow O-Header__navRow--subnav burger"> <div class="e-container"> <div class="O-Header__navRow__wrapper"> <nav aria-label="Breadcrumb"> <ol class="o-header__breadcrumbs"> <li> <div class="a-link -icon"><a href="/"><span>Home<i class="a-icon ui-ic-nosafe-lr-forward-small" title="Icon arrow right"></i></span></a></div> </li> <li> <div class="a-link -icon"><a href="/research/fields-of-expertise/"><span>Research<i class="a-icon ui-ic-nosafe-lr-forward-small" title="Icon arrow right"></i></span></a></div> </li> <li> <div class="a-link -icon"><a href="/research/fields-of-expertise/"><span>Fields Of Expertise</span> <span><i class="a-icon ui-ic-nosafe-lr-forward-small" title="Icon arrow right"></i></span></a></div> </li> <li aria-current="page"> Deep Learning </li> </ol> </nav> </div> </div> </div> </header> <main id="main" tabindex="-1"> <div class="M-Stage-Two" data-js-component="stage-two"> <div class="e-container"> <div class="M-Stage-Two__header"> <div class="M-Stage-Two__topbar"> <div class="M-Stage-Two__kicker">Our Fields Of Expertise </div> </div> <div class="M-Stage-Two__headline"> <div class="M-Stage-Two__headline-container a-text"> <h1>Deep Learning </h1> </div> </div> </div> </div> </div> <section class="M-TextTwo" aria-label="What Motivates Us"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>What Motivates Us</h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Deep learning refers to a modern class of machine learning systems that rely on multi-stage processing of data in neural networks. As a field, it has produced some of the most exciting advances in AI in recent years: breakthroughs in computer vision, language understanding and speech recognition have all had their roots in deep learning. This makes deep learning the key enabler for Bosch applications in fields like automated driving, robotics, or embedded AI. However, despite this promise, modern deep learning systems also have some deficiencies that make them ill-suited to tasks relevant to Bosch: the systems are compute and data-hungry, often brittle in real-world deployments, and typically unexplainable. Addressing these challenges motivates our team to create robust, safe, and efficient deep learning systems as part of continuously learning AIoT products.</p> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Our Approach"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Our Approach</h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>At Bosch, we are pushing forward the boundaries of deep learning to make these systems fully realizable within products. We create new methods of explainable and robust deep learning models that are ready to be validated and safely used in real-world environments. We develop novel training methods and tools to scale deep learning systems towards working efficiently on various types of embedded hardware. This way, we extend deep learning concepts from computer vision to the entire range of Bosch sensors. Data-efficient learning and generative models allow us to augment data sets with realistic variations of scenes, thereby reducing the efforts for data collection and labelling.</p> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Application"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Application</h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>The broad product portfolio of Bosch offers countless application fields for modern deep learning systems. Some of our core domains include perception for driving assistants and automated driving, robotics, health care, smart cameras, and many other applications of Bosch AIoT devices.</p> </div> </div> </div> </div> </section> <section aria-label="M_Navigation_Tabs"> <div class="m-frok-tab-navigation" data-js-component="tab-navigation"> <div class="e-container"> <div class="a-tab-navigation__wrapper" data-js-component="tab-buttons"> <div class="a-tab-navigation__gradients"></div> <ol class="a-tab-navigation" data-active="#tab-5033646-1" role="tablist"> <li class="a-tab-navigation__item" role="none"> <button aria-label="tab-5033646-1" id="tab-5033646-1" type="button" role="tab" class="a-tab-navigation__tab" data-frok-tab-identifier="tab-5033646-1" aria-controls="tab-5033646-1"> <span class="a-tab-navigation__tab-content"> <span class="a-tab-navigation__label">Specific Focus Area - Embedded Deep Learning</span> </span> </button> </li> <li class="a-tab-navigation__item" role="none"> <button aria-label="tab-5033646-2" id="tab-5033646-2" type="button" role="tab" class="a-tab-navigation__tab" data-frok-tab-identifier="tab-5033646-2" aria-controls="tab-5033646-2"> <span class="a-tab-navigation__tab-content"> <span class="a-tab-navigation__label">Specific Focus Area - Data Efficiency and Uncertainty in Deep Learning</span> </span> </button> </li> <li class="a-tab-navigation__item" role="none"> <button aria-label="tab-5033646-3" id="tab-5033646-3" type="button" role="tab" class="a-tab-navigation__tab" data-frok-tab-identifier="tab-5033646-3" aria-controls="tab-5033646-3"> <span class="a-tab-navigation__tab-content"> <span class="a-tab-navigation__label">Specific Focus Area - Explainable and Robust Deep Learning</span> </span> </button> </li> </ol> </div> <div class="e-container"> <div class="T-TabNavigationPage"> <div id="TAB-PANEL-5033646-tab-5033646-1" aria-labelledby="tab-5033646-1" role="tabpanel" class="a-tab-navigation-pane a-tab-navigation-pane--active" data-tab-name="tab-5033646-1" tabindex="0"> <section class="M-TextImageTwo" aria-label="Embedded Deep Learning "> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <h2>Embedded Deep Learning </h2> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp, /media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp, /media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp, /media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp, /media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_400x225.webp, /media/reserach/research_fields/bosch_animation_lokalisierung_04_complete_de_1920x1080_screenshot_06_road_signature_detection_v1_img_w1280_res_800x450.webp 2x" width="800" height="450" alt="Embedded Deep Learning "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p>Embedded deep learning aims to make the superior performance of deep learning algorithms for perception available for all types of sensors at Bosch while at the same time optimizing their computational footprint for deployment on embedded AI platforms and products. </p> <p>In order to make perception models run efficiently on resource-constrained hardware, we focus on methods such as hardware-aware neural architecture search. Those methods are key enablers for AI-powered products.</p> <p>Embedded deep learning is therefore being used in video, radar and ultrasonic sensors, amongst others, to improve their performance and efficiency. At Bosch, we collaborate closely between AI and hardware and domain experts to achieve the greatest benefits for embedded AI systems.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Use Case"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Use Case</h2> </div> </div> <div class="M-TextTwo__body"> <p class="M-TextTwo__intro-text"> Hardware-aware Neural Architecture Search Learning - Highly performant and efficient deep neural networks for embedded AI </p> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/embedded_learning1_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/embedded_learning1_res_400x225.webp, /media/reserach/research_fields/embedded_learning1_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_learning1_res_400x225.webp, /media/reserach/research_fields/embedded_learning1_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_learning1_res_400x225.webp, /media/reserach/research_fields/embedded_learning1_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_learning1_res_400x225.webp, /media/reserach/research_fields/embedded_learning1_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/embedded_learning1_res_400x225.webp, /media/reserach/research_fields/embedded_learning1_res_800x450.webp 2x" width="800" height="450" alt="deep learning "> </picture> </div> <figcaption class="A-Image__caption">Figure 1</figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="Introduction "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Introduction </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Imagine that your machine learning algorithm could automatically design the best possible neural network model for your task, and even adapt the network to run as efficiently as possible on the chosen hardware platform. Our research team has developed tools and algorithms that save time and costs for training your AI model, and deliver results that achieve great accuracy with orders of magnitude, smaller compute effort, and thus lower hardware costs and higher energy efficiency.</p> <p>Deep learning, despite its numerous performance improvements over classical machine learning techniques, builds computationally intensive models. This often precludes their use in embedding systems, where strict compute limits and performance guarantees are crucial. Our research on neural architecture search (NAS) aims for two goals:</p> <ul class="a-list a-list--dot"><li>Developing methods to extract the best possible performance from deep learning architectures for a specific compute budget, or tailored for a specific hardware platform.</li><li>Demonstrating the practicality of these methods by developing efficient deep learning architectures for domains of interest, such as deep networks, that enable the latest generation of video cameras or radar sensors.</li></ul> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Our Research "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Our Research </h2> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Tools for NAS </p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Differentiable and evolutionary NAS</strong></p> <p>We develop flexible and efficient re-usable Python tools that are building blocks for embedded AI applications in many domains. We integrate our latest research results, and specific optimizations for relevant hardware platforms.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/werkzeug_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Maximizing Efficiency</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/graph-2_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/graph-2_res_400x225.webp, /media/reserach/research_fields/graph-2_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-2_res_400x225.webp, /media/reserach/research_fields/graph-2_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-2_res_400x225.webp, /media/reserach/research_fields/graph-2_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-2_res_400x225.webp, /media/reserach/research_fields/graph-2_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/graph-2_res_400x225.webp, /media/reserach/research_fields/graph-2_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Hardware awareness leads to perfectly tailored solutions</strong></p> <p>In collaboration with hardware experts, we define models and cost functions that allow fine-tuning of deep learning models for various hardware platforms. The result are highly performant, and potentially orders of magnitude more efficient.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Pilot Use Cases</p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Demonstrating the benefits of NAS in embedded AI applications</strong></p> <p>NAS saves time to design network architectures for novel application fields and delivers near optimal solutions for complicated multi-objective optimization problems. Our tools are powerful, easy to apply, and can be integrated into continuously learning products.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp, /media/reserach/research_fields/research_field_01/rakete_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp, /media/reserach/research_fields/research_field_01/rakete_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp, /media/reserach/research_fields/research_field_01/rakete_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp, /media/reserach/research_fields/research_field_01/rakete_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/research_field_01/rakete_res_400x225.webp, /media/reserach/research_fields/research_field_01/rakete_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Ahead of the Curve</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/medaille-1_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/medaille-1_res_400x225.webp, /media/reserach/research_fields/medaille-1_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-1_res_400x225.webp, /media/reserach/research_fields/medaille-1_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-1_res_400x225.webp, /media/reserach/research_fields/medaille-1_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-1_res_400x225.webp, /media/reserach/research_fields/medaille-1_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/medaille-1_res_400x225.webp, /media/reserach/research_fields/medaille-1_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Cutting-edge research in neural architecture search</strong></p> <p>Our NAS tools and methods result from world-class research at BCAI and our academic collaborators at the University of Freiburg. We constantly challenge and advance the state-of-the-art in the field, publish at top-tier venues, and put methods to test in real-world Bosch applications.</p> </div> </div> </div> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp, /media/reserach/research_fields/embedded_deep_learning_2-2_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp, /media/reserach/research_fields/embedded_deep_learning_2-2_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp, /media/reserach/research_fields/embedded_deep_learning_2-2_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp, /media/reserach/research_fields/embedded_deep_learning_2-2_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/embedded_deep_learning_2-2_res_400x225.webp, /media/reserach/research_fields/embedded_deep_learning_2-2_res_800x450.webp 2x" width="800" height="450" alt="Deep Learning "> </picture> </div> <figcaption class="A-Image__caption">Figure 2</figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="References "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>References </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Elsken, T., Metzen, J. H., & Hutter, F. (2019). Neural architecture search: A survey. The Journal of Machine Learning Research, 20(1), 1997-2017. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://www.jmlr.org/papers/volume20/18-598/18-598.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Elsken, T., Metzen, J. H., & Hutter, F. (2018). Efficient Multi-Objective Neural Architecture Search via Lamarckian Evolution. ICLR. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/pdf/1804.09081.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Elsken, T., Zela, A., Metzen, J. H., Staffler, B., Brox, T., Valada, A., & Hutter, F. (2022). Neural Architecture Search for Dense Prediction Tasks in Computer Vision. arXiv preprint. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/pdf/2202.07242.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> </p> <p>Schorn, C., Elsken, T., Vogel, S., Runge, A., Guntoro, A., & Ascheid, G. (2020). Automated design of error-resilient and hardware-efficient deep neural networks. Neural Computing and Applications, 32(24), 18327-18345. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://link.springer.com/content/pdf/10.1007/s00521-020-04969-6.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> </div> </div> </div> </div> </section> </div> <div id="TAB-PANEL-5033646-tab-5033646-2" aria-labelledby="tab-5033646-2" role="tabpanel" class="a-tab-navigation-pane" data-tab-name="tab-5033646-2" tabindex="0"> <section class="M-TextImageTwo" aria-label="Data Efficiency and Uncertainty in Deep Learning"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <h2>Data Efficiency and Uncertainty in Deep Learning</h2> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/powerwall_male_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/powerwall_male_res_400x225.webp, /media/reserach/research_fields/powerwall_male_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/powerwall_male_res_400x225.webp, /media/reserach/research_fields/powerwall_male_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/powerwall_male_res_400x225.webp, /media/reserach/research_fields/powerwall_male_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/powerwall_male_res_400x225.webp, /media/reserach/research_fields/powerwall_male_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/powerwall_male_res_400x225.webp, /media/reserach/research_fields/powerwall_male_res_800x450.webp 2x" width="800" height="450" alt="Data Efficiency and Uncertainty in Deep Learning"> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p>Data is essential to achieve good performance and generalization of deep learning models. However, many restrictions in the data collection process often lead to the absence of sufficient amount of training data to enable adequate production performance.</p> <p>Therefore, data efficiency is important in deep learning since it enables the models to learn in complex domains without requiring large quantities of data. For this reason, we focus on synthetic data generation as well as unsupervised learning and domain adaptation techniques.</p> <p>A further challenge with deep learning models trained with limited data, but operating in an open world, is that they must be aware of the limits of their knowledge and the inherent uncertainty at the inference stage.</p> <p>To encounter that challenge, we are working on deep learning models that can make decisions under uncertainty and provide high-quality uncertainty quantification in order to increase safety of the systems and improve their out-of-distribution generalization.</p> <p>An example for application are deep learning models in the automotive domain that are robust to different corner cases like unusual weather or rare objects in the scene.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Use Case"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Use Case</h2> </div> </div> <div class="M-TextTwo__body"> <p class="M-TextTwo__intro-text"> Synthetic Data Generation and Augmentation - Enabling high performance and robustness of AI models with less efforts on data collection </p> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_res_800x450.webp 2x" width="800" height="450" alt="Data Efficiency and Uncertainty in Deep Learning "> </picture> </div> <figcaption class="A-Image__caption">To ensure safety, a model put into production must generalize well across different scenarios and be robust to various corner-cases. To enable it, these rare cases must be sufficiently present in the training data. </figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="Introduction "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Introduction </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Imagine an AI model that generalizes well to different scenarios (e.g. rare objects in the scene) and being robust to various corner-cases (e.g. unusual weather). To enable this, rare corner cases must be sufficiently present in your training data. However, they are often extremely hard and costly to collect. One way to increase the representation of these rare samples in the training data is by synthesizing them using deep generative models, such as Generative Adversarial Networks (GANs) or diffusion models.</p> <p>At BCAI, we are developing novel and high-performance models for data synthesis aiming to:</p> <ul class="a-list a-list--dot"><li> Reduce data collection and labelling costs by using synthetic data for augmentation,</li><li> Enable synthesis of realistically looking samples with a particular focus on the underrepresented samples in the collected data,</li><li> Improve overall performance of the downstream models and their robustness on rare cases by using synthetic data for training.</li></ul> </div> </div> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_400x225.webp, /media/reserach/research_fields/data_efficiency_and_uncertainty_in_deep_learning_2-3_res_800x450.webp 2x" width="800" height="450" alt="Data Efficiency and Uncertainty in Deep Learning "> </picture> </div> <figcaption class="A-Image__caption">Images synthesized by our OASIS model by manipulating latent directions, corresponding to day-to-night and cloudy-to-rainy scene properties.</figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="Our Research "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Our Research </h2> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Extra Data in the Loop for Free</p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Reducing the data collection and annotation costs</strong></p> <p>Data collection and annotation costs can be significantly reduced by using synthetic data generation tools for augmentation. Large amount of high quality training data can be synthesized for free and used to improve the performance of downstream models.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/geld_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/geld_res_400x225.webp, /media/reserach/research_fields/geld_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/geld_res_400x225.webp, /media/reserach/research_fields/geld_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/geld_res_400x225.webp, /media/reserach/research_fields/geld_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/geld_res_400x225.webp, /media/reserach/research_fields/geld_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/geld_res_400x225.webp, /media/reserach/research_fields/geld_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Dealing with Limited Data</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/batterie_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/batterie_res_400x225.webp, /media/reserach/research_fields/batterie_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/batterie_res_400x225.webp, /media/reserach/research_fields/batterie_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/batterie_res_400x225.webp, /media/reserach/research_fields/batterie_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/batterie_res_400x225.webp, /media/reserach/research_fields/batterie_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/batterie_res_400x225.webp, /media/reserach/research_fields/batterie_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Learning from one or a few data samples</strong></p> <p>Our algorithms also focus on use cases where training data is extremely limited, and only one or a few samples are available. Synthetic data generation methods can augment these samples with non-trivial transformations, boosting the performance of the downstream application models.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Improved Robustness</p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Unlocking new potential by learning from synthetic data</strong></p> <p>Our developed methods enable synthesis of realistically looking outlier and corner cases as well as synthesis of the underrepresented samples in the dataset. Those samples are crucial to improve the robustness of the model and often very difficult or even impossible to further collect.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/graph-3_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/graph-3_res_400x225.webp, /media/reserach/research_fields/graph-3_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-3_res_400x225.webp, /media/reserach/research_fields/graph-3_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-3_res_400x225.webp, /media/reserach/research_fields/graph-3_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-3_res_400x225.webp, /media/reserach/research_fields/graph-3_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/graph-3_res_400x225.webp, /media/reserach/research_fields/graph-3_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Ahead of the Curve</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/medaille-2_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/medaille-2_res_400x225.webp, /media/reserach/research_fields/medaille-2_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-2_res_400x225.webp, /media/reserach/research_fields/medaille-2_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-2_res_400x225.webp, /media/reserach/research_fields/medaille-2_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/medaille-2_res_400x225.webp, /media/reserach/research_fields/medaille-2_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/medaille-2_res_400x225.webp, /media/reserach/research_fields/medaille-2_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Benefitting from cutting-edge research on synthetic data generation</strong></p> <p>Our research team is continuously producing novel methods for synthetic data generation. We constantly challenge and advance the state-of-the-art in the field, publish at top-tier venues, and put methods to test in various real-world Bosch applications.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="References "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>References </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Zhang, D., & Khoreva, A. (2019). Progressive Augmentation of GANs. NeurIPS. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://proceedings.neurips.cc/paper/2019/hash/2e2c080d5490760af59d0baf5acbb84e-Abstract.html"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/boschresearch/PA-GAN"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Schoenfeld, E., Schiele, B., & Khoreva, A. (2020). A U-Net Based Discriminator for Generative Adversarial Networks. CVPR. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://openaccess.thecvf.com/content_CVPR_2020/papers/Schonfeld_A_U-Net_Based_Discriminator_for_Generative_Adversarial_Networks_CVPR_2020_paper.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/boschresearch/unetgan"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Video] opens external site" rel="noopener, noreferrer" href="https://www.youtube.com/watch?v=BR9C4p3W9vw"><span>[Video]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Schoenfeld, E., Sushko, V., Zhang, D., Gall, J., Schiele, B., Khoreva, A. (2021). You Only Need Adversarial Supervision for Semantic Image Synthesis. ICLR. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://openreview.net/pdf?id=yvQKLaqNE6M"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/boschresearch/OASIS"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Video] opens external site" rel="noopener, noreferrer" href="https://www.youtube.com/watch?v=vUm6vurIwyM"><span>[Video]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Sushko, V., Gall, J., Khoreva, A. (2021). One-Shot GAN: Learning to Generate Samples from Single Images and Videos. CVPR Workshops. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://openaccess.thecvf.com/content/CVPR2021W/LLID/papers/Sushko_One-Shot_GAN_Learning_To_Generate_Samples_From_Single_Images_and_CVPRW_2021_paper.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/boschresearch/one-shot-synthesis"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Video] opens external site" rel="noopener, noreferrer" href="https://www.youtube.com/watch?v=nsuC2sQvGfk"><span>[Video]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> </div> </div> </div> </div> </section> </div> <div id="TAB-PANEL-5033646-tab-5033646-3" aria-labelledby="tab-5033646-3" role="tabpanel" class="a-tab-navigation-pane" data-tab-name="tab-5033646-3" tabindex="0"> <section class="M-TextImageTwo" aria-label="Explainable and Robust Deep Learning"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <h2>Explainable and Robust Deep Learning</h2> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp, /media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp, /media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp, /media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp, /media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_400x225.webp, /media/reserach/research_fields/202102_ai_reutlingen_ae_manufacturing_scheduling_layer_2_res_800x450.webp 2x" width="800" height="450" alt="Explainable and Robust Deep Learning"> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p>Intelligent autonomous systems need to understand their environment – even if this environment changes over time and looks unlike what has been encountered previously. Moreover, providing an understanding of how and when the systems work reliably and robust can increase trust of users.</p> <p>We put a special emphasis on designing deep learning-based approaches for complex perception tasks such as multi-task perception, where for instance several object detection and semantic segmentation tasks are handled by a single neural network. Moreover, we are part of a larger team at Bosch exploring deep learning-based approaches for tracking objects over time and for localizing them in the 3D-world using multiple views of a scene.</p> <p>Besides providing high-performing solutions for perception problems, we also develop tools for quantifying and improving robustness, explainability, and calibration of neural networks. </p> <p>The main use cases of our perception systems are in domains like driver assistance, automated driving, or video surveillance.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Use Case"> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Use Case</h2> </div> </div> <div class="M-TextTwo__body"> <p class="M-TextTwo__intro-text"> Deep Learning for Temporal and Multi-View Fusion - Vision: Scalable, spatio-temporal consistent and real-time capable environment representation of vehicle surrounding </p> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion_res_800x450.webp 2x" width="800" height="450" alt="Deep Learning for Temporal and Multi-View Fusion"> </picture> </div> <figcaption class="A-Image__caption">A vehicle equipped with several cameras (left, fields of view in purple) generates a temporal sequence of per-camera frames (middle). We do research on AI-driven perception functionalities for generating an environment representation covering for instance information on vehicle locations and movements (right).</figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="Introduction "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Introduction </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Many intelligent systems need to be able to understand their surroundings based on sensors such as RGB cameras. Designing and training deep neural networks for perception tasks can be challenging as:</p> <ul class="a-list a-list--dot"><li> Information from various sensors need to aggregated to get a full understanding of a scene,</li><li> Sensory data need to be integrated over time to be able to track objects or estimate movements,</li><li> The size of neural architectures in embedded systems is limited by constraints on latency, energy consumption, and cost of hardware.</li></ul> <p>We address these challenges by developing neural architectures that are suited for multi-task perception, i.e. different perception tasks can be handled with the same network which reduces computation costs. In addition, we extend these networks to perform multiple view and temporal fusion which allows reasoning of 3D-geometry of the surrounding and tracking moving objects.</p> <p>A second challenge when designing neural networks for perception in highly automated systems is understanding their reliability and safety:</p> <ul class="a-list a-list--dot"><li> Can we understand the internal processing of sensory information in a perception system and use this understanding to increase trust in the system? </li><li> Can a trained network deal with situations that were not part of its training data (domain shift and corner cases)? How can we quantify and improve the robustness of a network to such unexpected data?</li><li> How can a neural network provide reliable estimates on the confidence of its predictions?</li><li> How can we validate a neural network for perception in order to deploy it in safety-critical use cases?</li></ul> <p>We aim on building tools that advance the state-of-the art of deep learning-based perception thus providing solutions for these challenging problems and questions.</p> </div> </div> </div> </div> </section> <section class="M-TextTwo" aria-label="Our Research "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>Our Research </h2> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Multi-Task Perception</p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Handling different perception tasks with a joint network</strong></p> <p>Neural networks can be used as models in different perception tasks such as object detection, semantic segmentation, or optical flow estimation. Using a shared neural backbone can reduce hardware requirements and latency.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/graph-4_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/graph-4_res_400x225.webp, /media/reserach/research_fields/graph-4_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-4_res_400x225.webp, /media/reserach/research_fields/graph-4_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-4_res_400x225.webp, /media/reserach/research_fields/graph-4_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/graph-4_res_400x225.webp, /media/reserach/research_fields/graph-4_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/graph-4_res_400x225.webp, /media/reserach/research_fields/graph-4_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Temporal and Multi-View Perception</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/werkzeug_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/werkzeug_res_400x225.webp, /media/reserach/research_fields/werkzeug_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Learning from temporal information of multiple sensors</strong></p> <p>Dynamic scenes are hard to capture in their entirety from a single static image. Integrating information over time and from sensors providing different views allows to achieve a much better understanding of the actual current state of the 3D-environment.</p> </div> </div> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Improved Robustness</p> </div> </div> <div class="M-TextImageTwo__body -ratio-8-to-4" data="leftAlignedSmall"> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Dealing with domain shift and adversarial attacks</strong></p> <p>Neural networks need to reliably process inputs that look unlike the typical inputs provided during their training, for instance those with a domain shift or those maliciously altered by an attacker. We devise architectures, training processes, and certification methods for robust neural perception.</p> </div> </div> </div> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp, /media/reserach/research_fields/research_field_01/medaille_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp, /media/reserach/research_fields/research_field_01/medaille_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp, /media/reserach/research_fields/research_field_01/medaille_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp, /media/reserach/research_fields/research_field_01/medaille_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/research_field_01/medaille_res_400x225.webp, /media/reserach/research_fields/research_field_01/medaille_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> </div> </div> </section> <section class="M-TextImageTwo" aria-label="M_Navigation_Tabs"> <div class="e-container"> <div class="a-component-header"> <div class="a-text"> <p>Interpretability and Validation</p> </div> </div> <div class="M-TextImageTwo__body -ratio-4-to-8" data="rightAlignedSmall"> <div class="M-TextImageTwo__imageWrapper"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/gluehbirne_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/gluehbirne_res_400x225.webp, /media/reserach/research_fields/gluehbirne_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/gluehbirne_res_400x225.webp, /media/reserach/research_fields/gluehbirne_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/gluehbirne_res_400x225.webp, /media/reserach/research_fields/gluehbirne_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/gluehbirne_res_400x225.webp, /media/reserach/research_fields/gluehbirne_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/gluehbirne_res_400x225.webp, /media/reserach/research_fields/gluehbirne_res_800x450.webp 2x" width="800" height="450" alt="Icon "> </picture> </div> </figure> </div> <div class="M-TextImageTwo__textWrapper"> <div class="M-TextImageTwo__body--description"> <div class="A-Text-RichText"> <p><strong>Understanding how and when neural networks work</strong></p> <p>Wide-spread adoption of neural networks for perception in sensitive or safety-critical domains requires understanding on how they operate internally and that they do not exploit spurious correlations that could hamper robustness and fairness. We develop novel approaches for explainability and validation of deep neural networks.</p> </div> </div> </div> </div> </div> </section> <ul class="M-ImageGroup__list"> <li class="M-ImageGroup__item"> <figure class="A-Image__figure" > <div class="A-Image__preloadWrapper A-Image__preloadWrapper--ratio16to9"> <div itemtype="https://schema.org/ImageObject" itemscope="" itemprop="image"> <meta content="https://www.bosch-ai.com/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp" itemprop="url" /> <meta content="100%" itemprop="width" /> <meta content="100%" itemprop="height" /> </div> <picture class="A-Image__picture" data-js-component="source-fallback"> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_800x450.webp 2x" media="(min-width: 1200px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_800x450.webp 2x" media="(min-width: 992px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_800x450.webp 2x" media="(min-width: 768px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <source data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_800x450.webp 2x" media="(max-width: 767px)" srcset="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw=="> <img class="A-Image A-Image--preload lazyload" src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" data-srcset="/media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_400x225.webp, /media/reserach/research_fields/deep_learning_for_temporal_and_multi-view_fusion-2_res_800x450.webp 2x" width="800" height="450" alt="Deep Learning for Temporal and Multi-View Fusion"> </picture> </div> <figcaption class="A-Image__caption">Shown is a sequence of semantic segmentations (top) and associated semantic grids (bottom, black areas correspond to unknown parts). Based on frame 1 and 2, a prediction of the semantic grid and associated uncertainty is generated and compared to a target.</figcaption> </figure> </li> </ul> <section class="M-TextTwo" aria-label="References "> <div class="e-container"> <div class="a-component-header -small"> <div class="a-text"> <h2>References </h2> </div> </div> <div class="M-TextTwo__body"> <div class="M-TextTwo__description"> <div class="A-Text-RichText"> <p>Eulig, E., Saranrittichai, P., Mummadi, C., Rambach, K., Beluch, W., Shi, X., & Fischer, V. (2021). DiagViB-6: A Diagnostic Benchmark Suite for Vision Models in the Presence of Shortcut and Generalization Opportunities. IEEE/CVF (ICCV). <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/pdf/2108.05779.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a><a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/boschresearch/diagvib-6"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Fuchs, F., Worrall, D., Fischer, V., & Welling, M. (2020). SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks. NeurIPS. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://proceedings.neurips.cc/paper/2020/hash/15231a7ce4ba789d13b722cc5c955834-Abstract.html"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Code] opens external site" rel="noopener, noreferrer" href="https://github.com/FabianFuchsML/se3-transformer-public"><span>[Code]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Video] opens external site" rel="noopener, noreferrer" href="https://slideslive.com/38931417/se3transformers-3d-rototranslation-equivariant-attention-networks"><span>[Video]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Mummadi, C., Subramaniam, R., Hutmacher, R., Vitay, J., Fischer, V., & Metzen J. (2021). Does enhanced shape bias improve neural network robustness to common corruptions? ICLR. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://openreview.net/forum?id=yUxUNaj2Sl"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a> <a target="_blank" aria-label="[Video] opens external site" rel="noopener, noreferrer" href="https://iclr.cc/virtual/2021/poster/3235"><span>[Video]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Mummadi, C., Hutmacher, R., Rambach, K., Levinkov, E., Brox, T., & Metzen J. (2021). Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformation. arXiv. <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/pdf/2106.14999.pdf"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Wong, E., Schneider, T., Schmitt, J., Schmidt, F. R., Kolter, J. Z. (2020). Neural Network Virtual Sensors for Fuel Injection Quantities with Provable Performance Specifications (IEEE IV). <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/abs/2007.00147"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Wong, E., Schmidt, F. R., Kolter, J. Z. (2019). Wasserstein Adversarial Examples via Projected Sinkhorn Iterations (ICML). <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/abs/1902.07906"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> <p>Wong, E., Schmidt, F. R., Metzen, J. H., Kolter, J. Z. (2018). Scaling Provable Adversarial Defenses. In Neural Information Processing Systems (NeurIPS). <a target="_blank" aria-label="[PDF] opens external site" rel="noopener, noreferrer" href="https://arxiv.org/abs/1805.12514"><span>[PDF]<i class="a-icon ui-ic-nosafe-lr-externallink" title="externallink"></i></span></a></p> </div> </div> </div> </div> </section> </div> </div> </div> </div> </div> </section> <section class="O-Strategic-Teaser-Group" aria-label="Decision Teaser"> <div class="e-container"> <div class="O-Strategic-Teaser-Group__items-wrapper"> <div class="O-Strategic-Teaser-Group__item"> <div class="M-Strategic-Teaser a-tile -contrast"> <a class="M-Strategic-Teaser__link" target="_self" aria-label="Academic Collaborations Learn more" href="/research/collaborations/"><div class="M-Strategic-Teaser__content"><div class="M-Strategic-Teaser__title">Academic Collaborations</div><span>Learn</span> <span>more<i class="a-icon ui-ic-nosafe-lr-forward-small" title="Icon arrow right"></i></span></div></a> </div> </div> <div class="O-Strategic-Teaser-Group__item"> <div class="M-Strategic-Teaser a-tile -contrast"> <a class="M-Strategic-Teaser__link" target="_self" aria-label="Publications Learn more" href="/research/publications/"><div class="M-Strategic-Teaser__content"><div class="M-Strategic-Teaser__title">Publications</div><span>Learn</span> <span>more<i class="a-icon ui-ic-nosafe-lr-forward-small" title="Icon arrow right"></i></span></div></a> </div> </div> </div> </div> </section> </main> <footer class="o-footer" aria-label="Footer main"> <div class="o-footer__search-container -secondary"> <div class="e-container"> <div class="o-footer__search"> <div class="o-footer__search-cta"> Still looking for something? </div> <div data-js-component="SearchField" class="M-SearchField--static" data-land-mark="footer" data-submit-label="Search" data-placeholder="Search" data-label="Search" data-aria-label-search-form="Side wide" data-aria-label-search-icon="Search" data-aria-label-close-icon="Close" data-error-message="Something went wrong during data retrieval." data-no-results-message="There are no matches for your search query." data-autosuggestions-link="See all results" data-keep-search-form-open="false" data-boolean-disable-auto-search="false" data-execute-search="" data-containing-element-selector="footer" data-parse-term-from-url-on-init="true" data-search-url="/search.html" data-show-suggestions-element-selector=".o-footer"> </div> </div> </div> </div> <div class="e-container"> <div class="o-footer__top"> <div class="o-footer__claim">Get in touch</div> <ul class="o-footer__links"> <li><div class="a-link"><a target="_self" aria-label="General contact info" href="/contact/"><span>General contact info</span></a></div></li> <li><div class="a-link"><a target="_self" aria-label="Join us" href="/careers/"><span>Join us</span></a></div></li> </ul> <div class="o-footer__share"> <div class="M-SocialShareIcons M-SocialShareIcons--leftAligned"> <div class="M-SocialShareIcons__icons M-SocialShareIcons__icons--leftAligned"> <a itemprop="sameAs" class="M-SocialShareIcons__icon" target="_blank" aria-label="Linkedin opens external site" rel="noopener noreferrer" href="https://www.linkedin.com/company/bosch-center-for-artificial-intelligence-bcai"><i class="a-icon boschicon-bosch-ic-linkedin" title="This icon opens the share dialogue of LinkedIn."></i></a> </div> </div> </div> </div> <hr class="a-divider"/> <div class="o-footer__bottom"> <ul class="o-footer__links"> <li><div class="a-link"><a target="_self" aria-label="Corporate Information" href="/corporate-information/corporate-information/"><span>Corporate Information</span></a></div></li> <li><div class="a-link"><a target="_self" aria-label="Legal notice" href="/corporate-information/legal-notice/"><span>Legal notice</span></a></div></li> <li><div class="a-link"><a target="_self" aria-label="Data Protection Policy (Privacy Policy)" href="/corporate-information/privacy-policy/"><span>Data Protection Policy (Privacy Policy)</span></a></div></li> </ul> <div class="o-footer__copyright"> © 2024 Robert Bosch GmbH </div> <button type="button" class="a-button a-button--integrated -without-label o-footer__back-to-top" title="back to top" aria-label="back to top" data-js-component="button-back-to-top" > <i class="a-icon boschicon-bosch-ic-up a-button__icon" title="Icon arrow up"></i> </button> </div> </div> </footer> <script defer src="/media/tech/scripts/main.js?ver=1716558076529"></script> </body> </html>