{"id":4667,"date":"2019-09-05T15:37:06","date_gmt":"2019-09-05T15:37:06","guid":{"rendered":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/?page_id=4667"},"modified":"2019-09-11T15:19:04","modified_gmt":"2019-09-11T15:19:04","slug":"trinn","status":"publish","type":"page","link":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/","title":{"rendered":"Transport Theory for Invertible Neural Networks (TRINN)"},"content":{"rendered":"<p>Deep Neural Networks have achieved remarkable progress in data analysis in a broad range of topics such as object recognition, machine translation and medical diagnostics.<\/p>\n<p><div id=\"attachment_4683\" style=\"width: 730px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird.png\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-4683\" class=\"wp-image-4683 size-large\" src=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-1024x255.png\" alt=\"\" width=\"720\" height=\"179\" srcset=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-1024x255.png 1024w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-300x75.png 300w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-768x191.png 768w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird.png 1506w\" sizes=\"auto, (max-width: 720px) 100vw, 720px\" \/><\/a><p id=\"caption-attachment-4683\" class=\"wp-caption-text\">An invertible neural net (INN) was trained to add colour to grayscale images. Which one is the original image? Answer at the bottom of the page. Image adapted from: [1]<\/p><\/div>Despite the success stories, findings are being published highlighting the current short-comings of this central and intensive research.\u00a0Classical Machine Learning theory can neither explain the success nor the deficiencies of deep networks. A new theory that could serve as the foundation in constructing network architectures and validating them does not exist. This is a major showstopper before the powerful technology can be safely used in the context of important or even security relevant applications.<\/p>\n<p><div id=\"attachment_4684\" style=\"width: 730px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation.jpg\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-4684\" class=\"wp-image-4684 size-large\" src=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation-1024x515.jpg\" alt=\"\" width=\"720\" height=\"362\" srcset=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation-1024x515.jpg 1024w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation-300x151.jpg 300w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation-768x386.jpg 768w, https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/street-segmentation.jpg 1577w\" sizes=\"auto, (max-width: 720px) 100vw, 720px\" \/><\/a><p id=\"caption-attachment-4684\" class=\"wp-caption-text\">Critical application for neural networks: Decide where a car can drive or not. Confidence estimation is critical to make the right decisions. Image adapted from: [2]<\/p><\/div>Our project aims to find an answer to the following fundamental questions:<\/p>\n<ul>\n<li>How can we design deep networks under given constraints?<\/li>\n<li>(How) Can we be confident in the answers given by a network?<\/li>\n<li>Can we explain the behaviour of a network?<\/li>\n<\/ul>\n<p>The questions are tackled using the new unconventional framework of transport theory: We treat the input and output of a network as probability distributions. This enables an easier mathematical and algorithmical approach. It is realised in the form of\u00a0<a href=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/inverse-problems-invertible-neural-networks\/\"><em>Invertible Neural Networks (INNs)<\/em><\/a>.<\/p>\n<h2>Literature<\/h2>\n<p>[1] L. Ardizzone, J. Kruse, C. L\u00fcth, C. Rother, and U. K\u00f6the. Guided image generation with conditional invertible neural networks. <a href=\"https:\/\/arxiv.org\/abs\/1907.02392\">arXiv:1907.02392<\/a>, 2019.<\/p>\n<p>[2]\u00a0Kendall, A., and Gal, Y.. What uncertainties do we need in bayesian deep learning for computer vision?. In <a href=\"http:\/\/papers.nips.cc\/paper\/7141-what-uncertainties-do-we-need-in-bayesian-deep-learning-for-computer-vision.pdf\">Advances in neural information processing systems 2017<\/a> (pp. 5574-5584).<\/p>\n<p><span style=\"font-size: 10pt;\">The original bird is the one all the way to the right.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Deep Neural Networks have achieved remarkable progress in data analysis in a broad range of topics such as object recognition, machine translation and medical diagnostics. Despite the success stories, findings are being published highlighting the current short-comings of this central and intensive research.\u00a0Classical Machine Learning theory can neither explain the success nor the deficiencies of [&hellip;]<\/p>\n","protected":false},"author":6,"featured_media":0,"parent":3632,"menu_order":2,"comment_status":"closed","ping_status":"closed","template":"","meta":{"inline_featured_image":false,"footnotes":""},"class_list":["post-4667","page","type-page","status-publish","hentry","post"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.4 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg\" \/>\n<meta property=\"og:description\" content=\"Deep Neural Networks have achieved remarkable progress in data analysis in a broad range of topics such as object recognition, machine translation and medical diagnostics. Despite the success stories, findings are being published highlighting the current short-comings of this central and intensive research.\u00a0Classical Machine Learning theory can neither explain the success nor the deficiencies of [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/\" \/>\n<meta property=\"og:site_name\" content=\"Computer Vision and Learning Lab Heidelberg\" \/>\n<meta property=\"article:modified_time\" content=\"2019-09-11T15:19:04+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-1024x255.png\" \/>\n<meta name=\"twitter:label1\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data1\" content=\"2 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/\",\"url\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/\",\"name\":\"Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/wp-content\\\/uploads\\\/2019\\\/09\\\/bird-1024x255.png\",\"datePublished\":\"2019-09-05T15:37:06+00:00\",\"dateModified\":\"2019-09-11T15:19:04+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/#primaryimage\",\"url\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/wp-content\\\/uploads\\\/2019\\\/09\\\/bird.png\",\"contentUrl\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/wp-content\\\/uploads\\\/2019\\\/09\\\/bird.png\",\"width\":1506,\"height\":375,\"caption\":\"An invertible neural net (INN) was trained to colour images. Which one is the right?\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/trinn\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Research\",\"item\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Machine Learning\",\"item\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/research\\\/machine-learning\\\/\"},{\"@type\":\"ListItem\",\"position\":4,\"name\":\"Transport Theory for Invertible Neural Networks (TRINN)\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/#website\",\"url\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/\",\"name\":\"Computer Vision and Learning Lab Heidelberg\",\"description\":\"Computer Vision and Learning Lab Heidelberg\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/hci.iwr.uni-heidelberg.de\\\/vislearn\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/","og_locale":"en_US","og_type":"article","og_title":"Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg","og_description":"Deep Neural Networks have achieved remarkable progress in data analysis in a broad range of topics such as object recognition, machine translation and medical diagnostics. Despite the success stories, findings are being published highlighting the current short-comings of this central and intensive research.\u00a0Classical Machine Learning theory can neither explain the success nor the deficiencies of [&hellip;]","og_url":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/","og_site_name":"Computer Vision and Learning Lab Heidelberg","article_modified_time":"2019-09-11T15:19:04+00:00","og_image":[{"url":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-1024x255.png","type":"","width":"","height":""}],"twitter_misc":{"Est. reading time":"2 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/","url":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/","name":"Transport Theory for Invertible Neural Networks (TRINN) - Computer Vision and Learning Lab Heidelberg","isPartOf":{"@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/#website"},"primaryImageOfPage":{"@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/#primaryimage"},"image":{"@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/#primaryimage"},"thumbnailUrl":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird-1024x255.png","datePublished":"2019-09-05T15:37:06+00:00","dateModified":"2019-09-11T15:19:04+00:00","breadcrumb":{"@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/#primaryimage","url":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird.png","contentUrl":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-content\/uploads\/2019\/09\/bird.png","width":1506,"height":375,"caption":"An invertible neural net (INN) was trained to colour images. Which one is the right?"},{"@type":"BreadcrumbList","@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/trinn\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/"},{"@type":"ListItem","position":2,"name":"Research","item":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/"},{"@type":"ListItem","position":3,"name":"Machine Learning","item":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/research\/machine-learning\/"},{"@type":"ListItem","position":4,"name":"Transport Theory for Invertible Neural Networks (TRINN)"}]},{"@type":"WebSite","@id":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/#website","url":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/","name":"Computer Vision and Learning Lab Heidelberg","description":"Computer Vision and Learning Lab Heidelberg","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"}]}},"_links":{"self":[{"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/pages\/4667","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/comments?post=4667"}],"version-history":[{"count":6,"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/pages\/4667\/revisions"}],"predecessor-version":[{"id":5236,"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/pages\/4667\/revisions\/5236"}],"up":[{"embeddable":true,"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/pages\/3632"}],"wp:attachment":[{"href":"https:\/\/hci.iwr.uni-heidelberg.de\/vislearn\/wp-json\/wp\/v2\/media?parent=4667"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}