Search | arXiv e-print repository

<!DOCTYPE html> <html lang="en"> <head> <meta charset="utf-8"/> <meta name="viewport" content="width=device-width, initial-scale=1"/>  <link rel="apple-touch-icon" sizes="180x180" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/apple-touch-icon.png"> <link rel="icon" type="image/png" sizes="32x32" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/favicon-32x32.png"> <link rel="icon" type="image/png" sizes="16x16" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/favicon-16x16.png"> <link rel="manifest" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/site.webmanifest"> <link rel="mask-icon" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/safari-pinned-tab.svg" color="#b31b1b"> <link rel="shortcut icon" href="https://static.arxiv.org/static/base/1.0.0a5/images/icons/favicon.ico"> <meta name="msapplication-TileColor" content="#b31b1b"> <meta name="msapplication-config" content="images/icons/browserconfig.xml"> <meta name="theme-color" content="#b31b1b">  <title>Search | arXiv e-print repository</title> <script defer src="https://static.arxiv.org/static/base/1.0.0a5/fontawesome-free-5.11.2-web/js/all.js"></script> <link rel="stylesheet" href="https://static.arxiv.org/static/base/1.0.0a5/css/arxivstyle.css" /> <script type="text/x-mathjax-config"> MathJax.Hub.Config({ messageStyle: "none", extensions: ["tex2jax.js"], jax: ["input/TeX", "output/HTML-CSS"], tex2jax: { inlineMath: [ ['$','$'], ["\$","\$"] ], displayMath: [ ['$$','$$'], ["\\[","\\]"] ], processEscapes: true, ignoreClass: '.*', processClass: 'mathjax.*' }, TeX: { extensions: ["AMSmath.js", "AMSsymbols.js", "noErrors.js"], noErrors: { inlineDelimiters: ["$","$"], multiLine: false, style: { "font-size": "normal", "border": "" } } }, "HTML-CSS": { availableFonts: ["TeX"] } }); </script> <script src='//static.arxiv.org/MathJax-2.7.3/MathJax.js'></script> <script src="https://static.arxiv.org/static/base/1.0.0a5/js/notification.js"></script> <link rel="stylesheet" href="https://static.arxiv.org/static/search/0.5.6/css/bulma-tooltip.min.css" /> <link rel="stylesheet" href="https://static.arxiv.org/static/search/0.5.6/css/search.css" /> <script src="https://code.jquery.com/jquery-3.2.1.slim.min.js" integrity="sha256-k2WSCIexGzOj3Euiig+TlR8gA0EmPjuc79OEeY5L45g=" crossorigin="anonymous"></script> <script src="https://static.arxiv.org/static/search/0.5.6/js/fieldset.js"></script> <style> radio#cf-customfield_11400 { display: none; } </style> </head> <body> <header><a href="#main-container" class="is-sr-only">Skip to main content</a>  <div class="attribution level is-marginless" role="banner"> <div class="level-left"> <a class="level-item" href="https://cornell.edu/"><img src="https://static.arxiv.org/static/base/1.0.0a5/images/cornell-reduced-white-SMALL.svg" alt="Cornell University" width="200" aria-label="logo" /></a> </div> <div class="level-right is-marginless"><p class="sponsors level-item is-marginless"><span id="support-ack-url">We gratefully acknowledge support from<br /> the Simons Foundation, <a href="https://info.arxiv.org/about/ourmembers.html">member institutions</a>, and all contributors. <a href="https://info.arxiv.org/about/donate.html">Donate</a></span></p></div> </div>  <div class="identity level is-marginless"> <div class="level-left"> <div class="level-item"> <a class="arxiv" href="https://arxiv.org/" aria-label="arxiv-logo"> <img src="https://static.arxiv.org/static/base/1.0.0a5/images/arxiv-logo-one-color-white.svg" aria-label="logo" alt="arxiv logo" width="85" style="width:85px;"/> </a> </div> </div> <div class="search-block level-right"> <form class="level-item mini-search" method="GET" action="https://arxiv.org/search"> <div class="field has-addons"> <div class="control"> <input class="input is-small" type="text" name="query" placeholder="Search..." aria-label="Search term or terms" /> <p class="help"><a href="https://info.arxiv.org/help">Help</a> | <a href="https://arxiv.org/search/advanced">Advanced Search</a></p> </div> <div class="control"> <div class="select is-small"> <select name="searchtype" aria-label="Field to search"> <option value="all" selected="selected">All fields</option> <option value="title">Title</option> <option value="author">Author</option> <option value="abstract">Abstract</option> <option value="comments">Comments</option> <option value="journal_ref">Journal reference</option> <option value="acm_class">ACM classification</option> <option value="msc_class">MSC classification</option> <option value="report_num">Report number</option> <option value="paper_id">arXiv identifier</option> <option value="doi">DOI</option> <option value="orcid">ORCID</option> <option value="author_id">arXiv author ID</option> <option value="help">Help pages</option> <option value="full_text">Full text</option> </select> </div> </div> <input type="hidden" name="source" value="header"> <button class="button is-small is-cul-darker">Search</button> </div> </form> </div> </div>  <div class="container"> <div class="user-tools is-size-7 has-text-right has-text-weight-bold" role="navigation" aria-label="User menu"> <a href="https://arxiv.org/login">Login</a> </div> </div> </header> <main class="container" id="main-container"> <div class="level is-marginless"> <div class="level-left"> <h1 class="title is-clearfix"> Showing 1–50 of 579 results for author: <span class="mathjax">Bengio, Y</span> </h1> </div> <div class="level-right is-hidden-mobile">  <span class="help" style="display: inline-block;"><a href="https://github.com/arXiv/arxiv-search/releases">Search v0.5.6 released 2020-02-24</a>  </span> </div> </div> <div class="content"> <form method="GET" action="/search/cs" aria-role="search"> Searching in archive <strong>cs</strong>. <a href="/search/?searchtype=author&query=Bengio%2C+Y">Search in all archives.</a> <div class="field has-addons-tablet"> <div class="control is-expanded"> <label for="query" class="hidden-label">Search term or terms</label> <input class="input is-medium" id="query" name="query" placeholder="Search term..." type="text" value="Bengio, Y"> </div> <div class="select control is-medium"> <label class="is-hidden" for="searchtype">Field</label> <select class="is-medium" id="searchtype" name="searchtype"><option value="all">All fields</option><option value="title">Title</option><option selected value="author">Author(s)</option><option value="abstract">Abstract</option><option value="comments">Comments</option><option value="journal_ref">Journal reference</option><option value="acm_class">ACM classification</option><option value="msc_class">MSC classification</option><option value="report_num">Report number</option><option value="paper_id">arXiv identifier</option><option value="doi">DOI</option><option value="orcid">ORCID</option><option value="license">License (URI)</option><option value="author_id">arXiv author ID</option><option value="help">Help pages</option><option value="full_text">Full text</option></select> </div> <div class="control"> <button class="button is-link is-medium">Search</button> </div> </div> <div class="field"> <div class="control is-size-7"> <label class="radio"> <input checked id="abstracts-0" name="abstracts" type="radio" value="show"> Show abstracts </label> <label class="radio"> <input id="abstracts-1" name="abstracts" type="radio" value="hide"> Hide abstracts </label> </div> </div> <div class="is-clearfix" style="height: 2.5em"> <div class="is-pulled-right"> <a href="/search/advanced?terms-0-term=Bengio%2C+Y&terms-0-field=author&size=50&order=-announced_date_first">Advanced Search</a> </div> </div> <input type="hidden" name="order" value="-announced_date_first"> <input type="hidden" name="size" value="50"> </form> <div class="level breathe-horizontal"> <div class="level-left"> <form method="GET" action="/search/"> <div style="display: none;"> <select id="searchtype" name="searchtype"><option value="all">All fields</option><option value="title">Title</option><option selected value="author">Author(s)</option><option value="abstract">Abstract</option><option value="comments">Comments</option><option value="journal_ref">Journal reference</option><option value="acm_class">ACM classification</option><option value="msc_class">MSC classification</option><option value="report_num">Report number</option><option value="paper_id">arXiv identifier</option><option value="doi">DOI</option><option value="orcid">ORCID</option><option value="license">License (URI)</option><option value="author_id">arXiv author ID</option><option value="help">Help pages</option><option value="full_text">Full text</option></select> <input id="query" name="query" type="text" value="Bengio, Y"> <ul id="abstracts"><li><input checked id="abstracts-0" name="abstracts" type="radio" value="show"> <label for="abstracts-0">Show abstracts</label></li><li><input id="abstracts-1" name="abstracts" type="radio" value="hide"> <label for="abstracts-1">Hide abstracts</label></li></ul> </div> <div class="box field is-grouped is-grouped-multiline level-item"> <div class="control"> <span class="select is-small"> <select id="size" name="size"><option value="25">25</option><option selected value="50">50</option><option value="100">100</option><option value="200">200</option></select> </span> <label for="size">results per page</label>. </div> <div class="control"> <label for="order">Sort results by</label> <span class="select is-small"> <select id="order" name="order"><option selected value="-announced_date_first">Announcement date (newest first)</option><option value="announced_date_first">Announcement date (oldest first)</option><option value="-submitted_date">Submission date (newest first)</option><option value="submitted_date">Submission date (oldest first)</option><option value="">Relevance</option></select> </span> </div> <div class="control"> <button class="button is-small is-link">Go</button> </div> </div> </form> </div> </div> <nav class="pagination is-small is-centered breathe-horizontal" role="navigation" aria-label="pagination"> <a href="" class="pagination-previous is-invisible">Previous </a> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=50" class="pagination-next" >Next </a> <ul class="pagination-list"> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=0" class="pagination-link is-current" aria-label="Goto page 1">1 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=50" class="pagination-link " aria-label="Page 2" aria-current="page">2 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=100" class="pagination-link " aria-label="Page 3" aria-current="page">3 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=150" class="pagination-link " aria-label="Page 4" aria-current="page">4 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=200" class="pagination-link " aria-label="Page 5" aria-current="page">5 </a> </li> <li><span class="pagination-ellipsis">…</span></li> </ul> </nav> <ol class="breathe-horizontal" start="1"> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2502.10236">arXiv:2502.10236</a> <span> [<a href="https://arxiv.org/pdf/2502.10236">pdf</a>, <a href="https://arxiv.org/format/2502.10236">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Shaping Inductive Bias in Diffusion Models through Frequency-Based Noise Control </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Jiralerspong%2C+T">Thomas Jiralerspong</a>, <a href="/search/cs?searchtype=author&query=Earnshaw%2C+B">Berton Earnshaw</a>, <a href="/search/cs?searchtype=author&query=Hartford%2C+J">Jason Hartford</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Scimeca%2C+L">Luca Scimeca</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2502.10236v1-abstract-short" style="display: inline;"> Diffusion Probabilistic Models (DPMs) are powerful generative models that have achieved unparalleled success in a number of generative tasks. In this work, we aim to build inductive biases into the training and sampling of diffusion models to better accommodate the target distribution of the data to model. For topologically structured data, we devise a frequency-based noising operator to purposefu… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.10236v1-abstract-full').style.display = 'inline'; document.getElementById('2502.10236v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2502.10236v1-abstract-full" style="display: none;"> Diffusion Probabilistic Models (DPMs) are powerful generative models that have achieved unparalleled success in a number of generative tasks. In this work, we aim to build inductive biases into the training and sampling of diffusion models to better accommodate the target distribution of the data to model. For topologically structured data, we devise a frequency-based noising operator to purposefully manipulate, and set, these inductive biases. We first show that appropriate manipulations of the noising forward process can lead DPMs to focus on particular aspects of the distribution to learn. We show that different datasets necessitate different inductive biases, and that appropriate frequency-based noise control induces increased generative performance compared to standard diffusion. Finally, we demonstrate the possibility of ignoring information at particular frequencies while learning. We show this in an image corruption and recovery task, where we train a DPM to recover the original target distribution after severe noise corruption. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.10236v1-abstract-full').style.display = 'none'; document.getElementById('2502.10236v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 14 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2025. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2502.07202">arXiv:2502.07202</a> <span> [<a href="https://arxiv.org/pdf/2502.07202">pdf</a>, <a href="https://arxiv.org/format/2502.07202">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Monte Carlo Tree Diffusion for System 2 Planning </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Yoon%2C+J">Jaesik Yoon</a>, <a href="/search/cs?searchtype=author&query=Cho%2C+H">Hyeonseo Cho</a>, <a href="/search/cs?searchtype=author&query=Baek%2C+D">Doojin Baek</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Ahn%2C+S">Sungjin Ahn</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2502.07202v1-abstract-short" style="display: inline;"> Diffusion models have recently emerged as a powerful tool for planning. However, unlike Monte Carlo Tree Search (MCTS)-whose performance naturally improves with additional test-time computation (TTC), standard diffusion-based planners offer only limited avenues for TTC scalability. In this paper, we introduce Monte Carlo Tree Diffusion (MCTD), a novel framework that integrates the generative stren… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.07202v1-abstract-full').style.display = 'inline'; document.getElementById('2502.07202v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2502.07202v1-abstract-full" style="display: none;"> Diffusion models have recently emerged as a powerful tool for planning. However, unlike Monte Carlo Tree Search (MCTS)-whose performance naturally improves with additional test-time computation (TTC), standard diffusion-based planners offer only limited avenues for TTC scalability. In this paper, we introduce Monte Carlo Tree Diffusion (MCTD), a novel framework that integrates the generative strength of diffusion models with the adaptive search capabilities of MCTS. Our method reconceptualizes denoising as a tree-structured process, allowing partially denoised plans to be iteratively evaluated, pruned, and refined. By selectively expanding promising trajectories while retaining the flexibility to revisit and improve suboptimal branches, MCTD achieves the benefits of MCTS such as controlling exploration-exploitation trade-offs within the diffusion framework. Empirical results on challenging long-horizon tasks show that MCTD outperforms diffusion baselines, yielding higher-quality solutions as TTC increases. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.07202v1-abstract-full').style.display = 'none'; document.getElementById('2502.07202v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 10 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2025. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">20 pages, 7 figures</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2502.06999">arXiv:2502.06999</a> <span> [<a href="https://arxiv.org/pdf/2502.06999">pdf</a>, <a href="https://arxiv.org/format/2502.06999">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Outsourced diffusion sampling: Efficient posterior inference in latent spaces of generative models </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Venkatraman%2C+S">Siddarth Venkatraman</a>, <a href="/search/cs?searchtype=author&query=Hasan%2C+M">Mohsin Hasan</a>, <a href="/search/cs?searchtype=author&query=Kim%2C+M">Minsu Kim</a>, <a href="/search/cs?searchtype=author&query=Scimeca%2C+L">Luca Scimeca</a>, <a href="/search/cs?searchtype=author&query=Sendera%2C+M">Marcin Sendera</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Berseth%2C+G">Glen Berseth</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2502.06999v1-abstract-short" style="display: inline;"> Any well-behaved generative model over a variable $\mathbf{x}$ can be expressed as a deterministic transformation of an exogenous ('outsourced') Gaussian noise variable $\mathbf{z}$: $\mathbf{x}=f_胃(\mathbf{z})$. In such a model (e.g., a VAE, GAN, or continuous-time flow-based model), sampling of the target variable $\mathbf{x} \sim p_胃(\mathbf{x})$ is straightforward, but sampling from a posterio… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.06999v1-abstract-full').style.display = 'inline'; document.getElementById('2502.06999v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2502.06999v1-abstract-full" style="display: none;"> Any well-behaved generative model over a variable $\mathbf{x}$ can be expressed as a deterministic transformation of an exogenous ('outsourced') Gaussian noise variable $\mathbf{z}$: $\mathbf{x}=f_胃(\mathbf{z})$. In such a model (e.g., a VAE, GAN, or continuous-time flow-based model), sampling of the target variable $\mathbf{x} \sim p_胃(\mathbf{x})$ is straightforward, but sampling from a posterior distribution of the form $p(\mathbf{x}\mid\mathbf{y}) \propto p_胃(\mathbf{x})r(\mathbf{x},\mathbf{y})$, where $r$ is a constraint function depending on an auxiliary variable $\mathbf{y}$, is generally intractable. We propose to amortize the cost of sampling from such posterior distributions with diffusion models that sample a distribution in the noise space ($\mathbf{z}$). These diffusion samplers are trained by reinforcement learning algorithms to enforce that the transformed samples $f_胃(\mathbf{z})$ are distributed according to the posterior in the data space ($\mathbf{x}$). For many models and constraints of interest, the posterior in the noise space is smoother than the posterior in the data space, making it more amenable to such amortized inference. Our method enables conditional sampling under unconditional GAN, (H)VAE, and flow-based priors, comparing favorably both with current amortized and non-amortized inference methods. We demonstrate the proposed outsourced diffusion sampling in several experiments with large pretrained prior models: conditional image generation, reinforcement learning with human feedback, and protein structure generation. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.06999v1-abstract-full').style.display = 'none'; document.getElementById('2502.06999v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 10 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2025. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2502.06323">arXiv:2502.06323</a> <span> [<a href="https://arxiv.org/pdf/2502.06323">pdf</a>, <a href="https://arxiv.org/format/2502.06323">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Materials Science">cond-mat.mtrl-sci</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> A physics-based data-driven model for CO$_2$ gas diffusion electrodes to drive automated laboratories </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Grega%2C+I">Ivan Grega</a>, <a href="/search/cs?searchtype=author&query=Therrien%2C+F">F茅lix Therrien</a>, <a href="/search/cs?searchtype=author&query=Soni%2C+A">Abhishek Soni</a>, <a href="/search/cs?searchtype=author&query=Ocean%2C+K">Karry Ocean</a>, <a href="/search/cs?searchtype=author&query=Dettelbach%2C+K">Kevan Dettelbach</a>, <a href="/search/cs?searchtype=author&query=Ahmadi%2C+R">Ribwar Ahmadi</a>, <a href="/search/cs?searchtype=author&query=Mokhtari%2C+M">Mehrdad Mokhtari</a>, <a href="/search/cs?searchtype=author&query=Berlinguette%2C+C+P">Curtis P. Berlinguette</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2502.06323v1-abstract-short" style="display: inline;"> The electrochemical reduction of atmospheric CO$_2$ into high-energy molecules with renewable energy is a promising avenue for energy storage that can take advantage of existing infrastructure especially in areas where sustainable alternatives to fossil fuels do not exist. Automated laboratories are currently being developed and used to optimize the composition and operating conditions of gas diff… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.06323v1-abstract-full').style.display = 'inline'; document.getElementById('2502.06323v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2502.06323v1-abstract-full" style="display: none;"> The electrochemical reduction of atmospheric CO$_2$ into high-energy molecules with renewable energy is a promising avenue for energy storage that can take advantage of existing infrastructure especially in areas where sustainable alternatives to fossil fuels do not exist. Automated laboratories are currently being developed and used to optimize the composition and operating conditions of gas diffusion electrodes (GDEs), the device in which this reaction takes place. Improving the efficiency of GDEs is crucial for this technology to become viable. Here we present a modeling framework to efficiently explore the high-dimensional parameter space of GDE designs in an active learning context. At the core of the framework is an uncertainty-aware physics model calibrated with experimental data. The model has the flexibility to capture various input parameter spaces and any carbon products which can be modeled with Tafel kinetics. It is interpretable, and a Gaussian process layer can capture deviations of real data from the function space of the physical model itself. We deploy the model in a simulated active learning setup with real electrochemical data gathered by the AdaCarbon automated laboratory and show that it can be used to efficiently traverse the multi-dimensional parameter space. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.06323v1-abstract-full').style.display = 'none'; document.getElementById('2502.06323v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 10 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2025. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">7 pages, 5 figures. Submitted to AI4Mat-ICLR2025 workshop</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2502.01341">arXiv:2502.01341</a> <span> [<a href="https://arxiv.org/pdf/2502.01341">pdf</a>, <a href="https://arxiv.org/format/2502.01341">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> </div> </div> <p class="title is-5 mathjax"> AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Understanding </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Masry%2C+A">Ahmed Masry</a>, <a href="/search/cs?searchtype=author&query=Rodriguez%2C+J+A">Juan A. Rodriguez</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+T">Tianyu Zhang</a>, <a href="/search/cs?searchtype=author&query=Wang%2C+S">Suyuchen Wang</a>, <a href="/search/cs?searchtype=author&query=Wang%2C+C">Chao Wang</a>, <a href="/search/cs?searchtype=author&query=Feizi%2C+A">Aarash Feizi</a>, <a href="/search/cs?searchtype=author&query=Suresh%2C+A+K">Akshay Kalkunte Suresh</a>, <a href="/search/cs?searchtype=author&query=Puri%2C+A">Abhay Puri</a>, <a href="/search/cs?searchtype=author&query=Jian%2C+X">Xiangru Jian</a>, <a href="/search/cs?searchtype=author&query=No%C3%ABl%2C+P">Pierre-Andr茅 No毛l</a>, <a href="/search/cs?searchtype=author&query=Madhusudhan%2C+S+T">Sathwik Tejaswi Madhusudhan</a>, <a href="/search/cs?searchtype=author&query=Pedersoli%2C+M">Marco Pedersoli</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+B">Bang Liu</a>, <a href="/search/cs?searchtype=author&query=Chapados%2C+N">Nicolas Chapados</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Hoque%2C+E">Enamul Hoque</a>, <a href="/search/cs?searchtype=author&query=Pal%2C+C">Christopher Pal</a>, <a href="/search/cs?searchtype=author&query=Laradji%2C+I+H">Issam H. Laradji</a>, <a href="/search/cs?searchtype=author&query=Vazquez%2C+D">David Vazquez</a>, <a href="/search/cs?searchtype=author&query=Taslakian%2C+P">Perouz Taslakian</a>, <a href="/search/cs?searchtype=author&query=Gella%2C+S">Spandana Gella</a>, <a href="/search/cs?searchtype=author&query=Rajeswar%2C+S">Sai Rajeswar</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2502.01341v1-abstract-short" style="display: inline;"> Aligning visual features with language embeddings is a key challenge in vision-language models (VLMs). The performance of such models hinges on having a good connector that maps visual features generated by a vision encoder to a shared embedding space with the LLM while preserving semantic similarity. Existing connectors, such as multilayer perceptrons (MLPs), often produce out-of-distribution or… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.01341v1-abstract-full').style.display = 'inline'; document.getElementById('2502.01341v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2502.01341v1-abstract-full" style="display: none;"> Aligning visual features with language embeddings is a key challenge in vision-language models (VLMs). The performance of such models hinges on having a good connector that maps visual features generated by a vision encoder to a shared embedding space with the LLM while preserving semantic similarity. Existing connectors, such as multilayer perceptrons (MLPs), often produce out-of-distribution or noisy inputs, leading to misalignment between the modalities. In this work, we propose a novel vision-text alignment method, AlignVLM, that maps visual features to a weighted average of LLM text embeddings. Our approach leverages the linguistic priors encoded by the LLM to ensure that visual features are mapped to regions of the space that the LLM can effectively interpret. AlignVLM is particularly effective for document understanding tasks, where scanned document images must be accurately mapped to their textual content. Our extensive experiments show that AlignVLM achieves state-of-the-art performance compared to prior alignment methods. We provide further analysis demonstrating improved vision-text feature alignment and robustness to noise. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2502.01341v1-abstract-full').style.display = 'none'; document.getElementById('2502.01341v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 3 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2025. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2501.17805">arXiv:2501.17805</a> <span> [<a href="https://arxiv.org/pdf/2501.17805">pdf</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> International AI Safety Report </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Mindermann%2C+S">S枚ren Mindermann</a>, <a href="/search/cs?searchtype=author&query=Privitera%2C+D">Daniel Privitera</a>, <a href="/search/cs?searchtype=author&query=Besiroglu%2C+T">Tamay Besiroglu</a>, <a href="/search/cs?searchtype=author&query=Bommasani%2C+R">Rishi Bommasani</a>, <a href="/search/cs?searchtype=author&query=Casper%2C+S">Stephen Casper</a>, <a href="/search/cs?searchtype=author&query=Choi%2C+Y">Yejin Choi</a>, <a href="/search/cs?searchtype=author&query=Fox%2C+P">Philip Fox</a>, <a href="/search/cs?searchtype=author&query=Garfinkel%2C+B">Ben Garfinkel</a>, <a href="/search/cs?searchtype=author&query=Goldfarb%2C+D">Danielle Goldfarb</a>, <a href="/search/cs?searchtype=author&query=Heidari%2C+H">Hoda Heidari</a>, <a href="/search/cs?searchtype=author&query=Ho%2C+A">Anson Ho</a>, <a href="/search/cs?searchtype=author&query=Kapoor%2C+S">Sayash Kapoor</a>, <a href="/search/cs?searchtype=author&query=Khalatbari%2C+L">Leila Khalatbari</a>, <a href="/search/cs?searchtype=author&query=Longpre%2C+S">Shayne Longpre</a>, <a href="/search/cs?searchtype=author&query=Manning%2C+S">Sam Manning</a>, <a href="/search/cs?searchtype=author&query=Mavroudis%2C+V">Vasilios Mavroudis</a>, <a href="/search/cs?searchtype=author&query=Mazeika%2C+M">Mantas Mazeika</a>, <a href="/search/cs?searchtype=author&query=Michael%2C+J">Julian Michael</a>, <a href="/search/cs?searchtype=author&query=Newman%2C+J">Jessica Newman</a>, <a href="/search/cs?searchtype=author&query=Ng%2C+K+Y">Kwan Yee Ng</a>, <a href="/search/cs?searchtype=author&query=Okolo%2C+C+T">Chinasa T. Okolo</a>, <a href="/search/cs?searchtype=author&query=Raji%2C+D">Deborah Raji</a>, <a href="/search/cs?searchtype=author&query=Sastry%2C+G">Girish Sastry</a>, <a href="/search/cs?searchtype=author&query=Seger%2C+E">Elizabeth Seger</a> , et al. (71 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2501.17805v1-abstract-short" style="display: inline;"> The first International AI Safety Report comprehensively synthesizes the current evidence on the capabilities, risks, and safety of advanced AI systems. The report was mandated by the nations attending the AI Safety Summit in Bletchley, UK. Thirty nations, the UN, the OECD, and the EU each nominated a representative to the report's Expert Advisory Panel. A total of 100 AI experts contributed, repr… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.17805v1-abstract-full').style.display = 'inline'; document.getElementById('2501.17805v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2501.17805v1-abstract-full" style="display: none;"> The first International AI Safety Report comprehensively synthesizes the current evidence on the capabilities, risks, and safety of advanced AI systems. The report was mandated by the nations attending the AI Safety Summit in Bletchley, UK. Thirty nations, the UN, the OECD, and the EU each nominated a representative to the report's Expert Advisory Panel. A total of 100 AI experts contributed, representing diverse perspectives and disciplines. Led by the report's Chair, these independent experts collectively had full discretion over the report's content. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.17805v1-abstract-full').style.display = 'none'; document.getElementById('2501.17805v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 29 January, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> January 2025. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2501.11183">arXiv:2501.11183</a> <span> [<a href="https://arxiv.org/pdf/2501.11183">pdf</a>, <a href="https://arxiv.org/ps/2501.11183">ps</a>, <a href="https://arxiv.org/format/2501.11183">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Cryptography and Security">cs.CR</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Can Safety Fine-Tuning Be More Principled? Lessons Learned from Cybersecurity </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Williams-King%2C+D">David Williams-King</a>, <a href="/search/cs?searchtype=author&query=Le%2C+L">Linh Le</a>, <a href="/search/cs?searchtype=author&query=Oberman%2C+A">Adam Oberman</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2501.11183v1-abstract-short" style="display: inline;"> As LLMs develop increasingly advanced capabilities, there is an increased need to minimize the harm that could be caused to society by certain model outputs; hence, most LLMs have safety guardrails added, for example via fine-tuning. In this paper, we argue the position that current safety fine-tuning is very similar to a traditional cat-and-mouse game (or arms race) between attackers and defender… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.11183v1-abstract-full').style.display = 'inline'; document.getElementById('2501.11183v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2501.11183v1-abstract-full" style="display: none;"> As LLMs develop increasingly advanced capabilities, there is an increased need to minimize the harm that could be caused to society by certain model outputs; hence, most LLMs have safety guardrails added, for example via fine-tuning. In this paper, we argue the position that current safety fine-tuning is very similar to a traditional cat-and-mouse game (or arms race) between attackers and defenders in cybersecurity. Model jailbreaks and attacks are patched with bandaids to target the specific attack mechanism, but many similar attack vectors might remain. When defenders are not proactively coming up with principled mechanisms, it becomes very easy for attackers to sidestep any new defenses. We show how current defenses are insufficient to prevent new adversarial jailbreak attacks, reward hacking, and loss of control problems. In order to learn from past mistakes in cybersecurity, we draw analogies with historical examples and develop lessons learned that can be applied to LLM safety. These arguments support the need for new and more principled approaches to designing safe models, which are architected for security from the beginning. We describe several such approaches from the AI literature. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.11183v1-abstract-full').style.display = 'none'; document.getElementById('2501.11183v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 19 January, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> January 2025. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">published at Neurips Safe Generative AI Workshop 2024</span> </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">ACM Class:</span> I.2.7; D.4.6 </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2501.08111">arXiv:2501.08111</a> <span> [<a href="https://arxiv.org/pdf/2501.08111">pdf</a>, <a href="https://arxiv.org/format/2501.08111">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> </div> </div> <p class="title is-5 mathjax"> EarthView: A Large Scale Remote Sensing Dataset for Self-Supervision </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Velazquez%2C+D">Diego Velazquez</a>, <a href="/search/cs?searchtype=author&query=L%C3%B3pez%2C+P+R">Pau Rodriguez L贸pez</a>, <a href="/search/cs?searchtype=author&query=Alonso%2C+S">Sergio Alonso</a>, <a href="/search/cs?searchtype=author&query=Gonfaus%2C+J+M">Josep M. Gonfaus</a>, <a href="/search/cs?searchtype=author&query=Gonzalez%2C+J">Jordi Gonzalez</a>, <a href="/search/cs?searchtype=author&query=Richarte%2C+G">Gerardo Richarte</a>, <a href="/search/cs?searchtype=author&query=Marin%2C+J">Javier Marin</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Lacoste%2C+A">Alexandre Lacoste</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2501.08111v1-abstract-short" style="display: inline;"> This paper presents EarthView, a comprehensive dataset specifically designed for self-supervision on remote sensing data, intended to enhance deep learning applications on Earth monitoring tasks. The dataset spans 15 tera pixels of global remote-sensing data, combining imagery from a diverse range of sources, including NEON, Sentinel, and a novel release of 1m spatial resolution data from Satellog… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.08111v1-abstract-full').style.display = 'inline'; document.getElementById('2501.08111v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2501.08111v1-abstract-full" style="display: none;"> This paper presents EarthView, a comprehensive dataset specifically designed for self-supervision on remote sensing data, intended to enhance deep learning applications on Earth monitoring tasks. The dataset spans 15 tera pixels of global remote-sensing data, combining imagery from a diverse range of sources, including NEON, Sentinel, and a novel release of 1m spatial resolution data from Satellogic. Our dataset provides a wide spectrum of image data with varying resolutions, harnessed from different sensors and organized coherently into an accessible HuggingFace dataset in parquet format. This data spans five years, from 2017 to 2022. Accompanying the dataset, we introduce EarthMAE, a tailored Masked Autoencoder, developed to tackle the distinct challenges of remote sensing data. Trained in a self-supervised fashion, EarthMAE effectively processes different data modalities such as hyperspectral, multispectral, topographical data, segmentation maps, and temporal structure. This model helps us show that pre-training on Satellogic data improves performance on downstream tasks. While there is still a gap to fill in MAE for heterogeneous data, we regard this innovative combination of an expansive, diverse dataset and a versatile model adapted for self-supervised learning as a stride forward in deep learning for Earth monitoring. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2501.08111v1-abstract-full').style.display = 'none'; document.getElementById('2501.08111v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 14 January, 2025; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> January 2025. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">2nd Workshop on Computer Vision for Earth Observation (CV4EO) Applications</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2412.07775">arXiv:2412.07775</a> <span> [<a href="https://arxiv.org/pdf/2412.07775">pdf</a>, <a href="https://arxiv.org/format/2412.07775">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> </div> </div> <p class="title is-5 mathjax"> Efficient Diversity-Preserving Diffusion Alignment via Gradient-Informed GFlowNets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Liu%2C+Z">Zhen Liu</a>, <a href="/search/cs?searchtype=author&query=Xiao%2C+T+Z">Tim Z. Xiao</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+W">Weiyang Liu</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+D">Dinghuai Zhang</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2412.07775v1-abstract-short" style="display: inline;"> While one commonly trains large diffusion models by collecting datasets on target downstream tasks, it is often desired to align and finetune pretrained diffusion models on some reward functions that are either designed by experts or learned from small-scale datasets. Existing methods for finetuning diffusion models typically suffer from lack of diversity in generated samples, lack of prior preser… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.07775v1-abstract-full').style.display = 'inline'; document.getElementById('2412.07775v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2412.07775v1-abstract-full" style="display: none;"> While one commonly trains large diffusion models by collecting datasets on target downstream tasks, it is often desired to align and finetune pretrained diffusion models on some reward functions that are either designed by experts or learned from small-scale datasets. Existing methods for finetuning diffusion models typically suffer from lack of diversity in generated samples, lack of prior preservation, and/or slow convergence in finetuning. Inspired by recent successes in generative flow networks (GFlowNets), a class of probabilistic models that sample with the unnormalized density of a reward function, we propose a novel GFlowNet method dubbed Nabla-GFlowNet (abbreviated as $\nabla$-GFlowNet), the first GFlowNet method that leverages the rich signal in reward gradients, together with an objective called $\nabla$-DB plus its variant residual $\nabla$-DB designed for prior-preserving diffusion alignment. We show that our proposed method achieves fast yet diversity- and prior-preserving alignment of Stable Diffusion, a large-scale text-conditioned image diffusion model, on different realistic reward functions. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.07775v1-abstract-full').style.display = 'none'; document.getElementById('2412.07775v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 10 December, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> December 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Technical Report (35 pages, 31 figures)</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2412.05282">arXiv:2412.05282</a> <span> [<a href="https://arxiv.org/pdf/2412.05282">pdf</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> International Scientific Report on the Safety of Advanced AI (Interim Report) </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Mindermann%2C+S">S枚ren Mindermann</a>, <a href="/search/cs?searchtype=author&query=Privitera%2C+D">Daniel Privitera</a>, <a href="/search/cs?searchtype=author&query=Besiroglu%2C+T">Tamay Besiroglu</a>, <a href="/search/cs?searchtype=author&query=Bommasani%2C+R">Rishi Bommasani</a>, <a href="/search/cs?searchtype=author&query=Casper%2C+S">Stephen Casper</a>, <a href="/search/cs?searchtype=author&query=Choi%2C+Y">Yejin Choi</a>, <a href="/search/cs?searchtype=author&query=Goldfarb%2C+D">Danielle Goldfarb</a>, <a href="/search/cs?searchtype=author&query=Heidari%2C+H">Hoda Heidari</a>, <a href="/search/cs?searchtype=author&query=Khalatbari%2C+L">Leila Khalatbari</a>, <a href="/search/cs?searchtype=author&query=Longpre%2C+S">Shayne Longpre</a>, <a href="/search/cs?searchtype=author&query=Mavroudis%2C+V">Vasilios Mavroudis</a>, <a href="/search/cs?searchtype=author&query=Mazeika%2C+M">Mantas Mazeika</a>, <a href="/search/cs?searchtype=author&query=Ng%2C+K+Y">Kwan Yee Ng</a>, <a href="/search/cs?searchtype=author&query=Okolo%2C+C+T">Chinasa T. Okolo</a>, <a href="/search/cs?searchtype=author&query=Raji%2C+D">Deborah Raji</a>, <a href="/search/cs?searchtype=author&query=Skeadas%2C+T">Theodora Skeadas</a>, <a href="/search/cs?searchtype=author&query=Tram%C3%A8r%2C+F">Florian Tram猫r</a>, <a href="/search/cs?searchtype=author&query=Adekanmbi%2C+B">Bayo Adekanmbi</a>, <a href="/search/cs?searchtype=author&query=Christiano%2C+P">Paul Christiano</a>, <a href="/search/cs?searchtype=author&query=Dalrymple%2C+D">David Dalrymple</a>, <a href="/search/cs?searchtype=author&query=Dietterich%2C+T+G">Thomas G. Dietterich</a>, <a href="/search/cs?searchtype=author&query=Felten%2C+E">Edward Felten</a>, <a href="/search/cs?searchtype=author&query=Fung%2C+P">Pascale Fung</a>, <a href="/search/cs?searchtype=author&query=Gourinchas%2C+P">Pierre-Olivier Gourinchas</a> , et al. (19 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2412.05282v1-abstract-short" style="display: inline;"> This is the interim publication of the first International Scientific Report on the Safety of Advanced AI. The report synthesises the scientific understanding of general-purpose AI -- AI that can perform a wide variety of tasks -- with a focus on understanding and managing its risks. A diverse group of 75 AI experts contributed to this report, including an international Expert Advisory Panel nomin… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.05282v1-abstract-full').style.display = 'inline'; document.getElementById('2412.05282v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2412.05282v1-abstract-full" style="display: none;"> This is the interim publication of the first International Scientific Report on the Safety of Advanced AI. The report synthesises the scientific understanding of general-purpose AI -- AI that can perform a wide variety of tasks -- with a focus on understanding and managing its risks. A diverse group of 75 AI experts contributed to this report, including an international Expert Advisory Panel nominated by 30 countries, the EU, and the UN. Led by the Chair, these independent experts collectively had full discretion over the report's content. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.05282v1-abstract-full').style.display = 'none'; document.getElementById('2412.05282v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 5 November, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> December 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Available under the open government license at https://www.gov.uk/government/publications/international-scientific-report-on-the-safety-of-advanced-ai</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2412.04626">arXiv:2412.04626</a> <span> [<a href="https://arxiv.org/pdf/2412.04626">pdf</a>, <a href="https://arxiv.org/format/2412.04626">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> </div> </div> <p class="title is-5 mathjax"> BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Rodriguez%2C+J">Juan Rodriguez</a>, <a href="/search/cs?searchtype=author&query=Jian%2C+X">Xiangru Jian</a>, <a href="/search/cs?searchtype=author&query=Panigrahi%2C+S+S">Siba Smarak Panigrahi</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+T">Tianyu Zhang</a>, <a href="/search/cs?searchtype=author&query=Feizi%2C+A">Aarash Feizi</a>, <a href="/search/cs?searchtype=author&query=Puri%2C+A">Abhay Puri</a>, <a href="/search/cs?searchtype=author&query=Kalkunte%2C+A">Akshay Kalkunte</a>, <a href="/search/cs?searchtype=author&query=Savard%2C+F">Fran莽ois Savard</a>, <a href="/search/cs?searchtype=author&query=Masry%2C+A">Ahmed Masry</a>, <a href="/search/cs?searchtype=author&query=Nayak%2C+S">Shravan Nayak</a>, <a href="/search/cs?searchtype=author&query=Awal%2C+R">Rabiul Awal</a>, <a href="/search/cs?searchtype=author&query=Massoud%2C+M">Mahsa Massoud</a>, <a href="/search/cs?searchtype=author&query=Abaskohi%2C+A">Amirhossein Abaskohi</a>, <a href="/search/cs?searchtype=author&query=Li%2C+Z">Zichao Li</a>, <a href="/search/cs?searchtype=author&query=Wang%2C+S">Suyuchen Wang</a>, <a href="/search/cs?searchtype=author&query=No%C3%ABl%2C+P">Pierre-Andr茅 No毛l</a>, <a href="/search/cs?searchtype=author&query=Richter%2C+M+L">Mats Leon Richter</a>, <a href="/search/cs?searchtype=author&query=Vadacchino%2C+S">Saverio Vadacchino</a>, <a href="/search/cs?searchtype=author&query=Agarwal%2C+S">Shubbam Agarwal</a>, <a href="/search/cs?searchtype=author&query=Biswas%2C+S">Sanket Biswas</a>, <a href="/search/cs?searchtype=author&query=Shanian%2C+S">Sara Shanian</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+Y">Ying Zhang</a>, <a href="/search/cs?searchtype=author&query=Bolger%2C+N">Noah Bolger</a>, <a href="/search/cs?searchtype=author&query=MacDonald%2C+K">Kurt MacDonald</a>, <a href="/search/cs?searchtype=author&query=Fauvel%2C+S">Simon Fauvel</a> , et al. (18 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2412.04626v1-abstract-short" style="display: inline;"> Multimodal AI has the potential to significantly enhance document-understanding tasks, such as processing receipts, understanding workflows, extracting data from documents, and summarizing reports. Code generation tasks that require long-structured outputs can also be enhanced by multimodality. Despite this, their use in commercial applications is often limited due to limited access to training da… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.04626v1-abstract-full').style.display = 'inline'; document.getElementById('2412.04626v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2412.04626v1-abstract-full" style="display: none;"> Multimodal AI has the potential to significantly enhance document-understanding tasks, such as processing receipts, understanding workflows, extracting data from documents, and summarizing reports. Code generation tasks that require long-structured outputs can also be enhanced by multimodality. Despite this, their use in commercial applications is often limited due to limited access to training data and restrictive licensing, which hinders open access. To address these limitations, we introduce BigDocs-7.5M, a high-quality, open-access dataset comprising 7.5 million multimodal documents across 30 tasks. We use an efficient data curation process to ensure our data is high-quality and license-permissive. Our process emphasizes accountability, responsibility, and transparency through filtering rules, traceable metadata, and careful content analysis. Additionally, we introduce BigDocs-Bench, a benchmark suite with 10 novel tasks where we create datasets that reflect real-world use cases involving reasoning over Graphical User Interfaces (GUI) and code generation from images. Our experiments show that training with BigDocs-Bench improves average performance up to 25.8% over closed-source GPT-4o in document reasoning and structured output tasks such as Screenshot2HTML or Image2Latex generation. Finally, human evaluations showed a preference for outputs from models trained on BigDocs over GPT-4o. This suggests that BigDocs can help both academics and the open-source community utilize and improve AI tools to enhance multimodal capabilities and document reasoning. The project is hosted at https://bigdocs.github.io . <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2412.04626v1-abstract-full').style.display = 'none'; document.getElementById('2412.04626v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 5 December, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> December 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">The project is hosted at https://bigdocs.github.io</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2411.02478">arXiv:2411.02478</a> <span> [<a href="https://arxiv.org/pdf/2411.02478">pdf</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Human-Computer Interaction">cs.HC</span> </div> </div> <p class="title is-5 mathjax"> Imagining and building wise machines: The centrality of AI metacognition </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Johnson%2C+S+G+B">Samuel G. B. Johnson</a>, <a href="/search/cs?searchtype=author&query=Karimi%2C+A">Amir-Hossein Karimi</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Chater%2C+N">Nick Chater</a>, <a href="/search/cs?searchtype=author&query=Gerstenberg%2C+T">Tobias Gerstenberg</a>, <a href="/search/cs?searchtype=author&query=Larson%2C+K">Kate Larson</a>, <a href="/search/cs?searchtype=author&query=Levine%2C+S">Sydney Levine</a>, <a href="/search/cs?searchtype=author&query=Mitchell%2C+M">Melanie Mitchell</a>, <a href="/search/cs?searchtype=author&query=Rahwan%2C+I">Iyad Rahwan</a>, <a href="/search/cs?searchtype=author&query=Sch%C3%B6lkopf%2C+B">Bernhard Sch枚lkopf</a>, <a href="/search/cs?searchtype=author&query=Grossmann%2C+I">Igor Grossmann</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2411.02478v1-abstract-short" style="display: inline;"> Recent advances in artificial intelligence (AI) have produced systems capable of increasingly sophisticated performance on cognitive tasks. However, AI systems still struggle in critical ways: unpredictable and novel environments (robustness), lack of transparency in their reasoning (explainability), challenges in communication and commitment (cooperation), and risks due to potential harmful actio… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2411.02478v1-abstract-full').style.display = 'inline'; document.getElementById('2411.02478v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2411.02478v1-abstract-full" style="display: none;"> Recent advances in artificial intelligence (AI) have produced systems capable of increasingly sophisticated performance on cognitive tasks. However, AI systems still struggle in critical ways: unpredictable and novel environments (robustness), lack of transparency in their reasoning (explainability), challenges in communication and commitment (cooperation), and risks due to potential harmful actions (safety). We argue that these shortcomings stem from one overarching failure: AI systems lack wisdom. Drawing from cognitive and social sciences, we define wisdom as the ability to navigate intractable problems - those that are ambiguous, radically uncertain, novel, chaotic, or computationally explosive - through effective task-level and metacognitive strategies. While AI research has focused on task-level strategies, metacognition - the ability to reflect on and regulate one's thought processes - is underdeveloped in AI systems. In humans, metacognitive strategies such as recognizing the limits of one's knowledge, considering diverse perspectives, and adapting to context are essential for wise decision-making. We propose that integrating metacognitive capabilities into AI systems is crucial for enhancing their robustness, explainability, cooperation, and safety. By focusing on developing wise AI, we suggest an alternative to aligning AI with specific human values - a task fraught with conceptual and practical difficulties. Instead, wise AI systems can thoughtfully navigate complex situations, account for diverse human values, and avoid harmful actions. We discuss potential approaches to building wise AI, including benchmarking metacognitive abilities and training AI systems to employ wise reasoning. Prioritizing metacognition in AI research will lead to systems that act not only intelligently but also wisely in complex, real-world situations. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2411.02478v1-abstract-full').style.display = 'none'; document.getElementById('2411.02478v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 4 November, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> November 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">26 pages, 1 figure, 2 tables</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.21154">arXiv:2410.21154</a> <span> [<a href="https://arxiv.org/pdf/2410.21154">pdf</a>, <a href="https://arxiv.org/format/2410.21154">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">stat.ML</span> </div> </div> <p class="title is-5 mathjax"> Trajectory Flow Matching with Applications to Clinical Time Series Modeling </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Zhang%2C+X">Xi Zhang</a>, <a href="/search/cs?searchtype=author&query=Pu%2C+Y">Yuan Pu</a>, <a href="/search/cs?searchtype=author&query=Kawamura%2C+Y">Yuki Kawamura</a>, <a href="/search/cs?searchtype=author&query=Loza%2C+A">Andrew Loza</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Shung%2C+D+L">Dennis L. Shung</a>, <a href="/search/cs?searchtype=author&query=Tong%2C+A">Alexander Tong</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.21154v2-abstract-short" style="display: inline;"> Modeling stochastic and irregularly sampled time series is a challenging problem found in a wide range of applications, especially in medicine. Neural stochastic differential equations (Neural SDEs) are an attractive modeling technique for this problem, which parameterize the drift and diffusion terms of an SDE with neural networks. However, current algorithms for training Neural SDEs require back… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.21154v2-abstract-full').style.display = 'inline'; document.getElementById('2410.21154v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.21154v2-abstract-full" style="display: none;"> Modeling stochastic and irregularly sampled time series is a challenging problem found in a wide range of applications, especially in medicine. Neural stochastic differential equations (Neural SDEs) are an attractive modeling technique for this problem, which parameterize the drift and diffusion terms of an SDE with neural networks. However, current algorithms for training Neural SDEs require backpropagation through the SDE dynamics, greatly limiting their scalability and stability. To address this, we propose Trajectory Flow Matching (TFM), which trains a Neural SDE in a simulation-free manner, bypassing backpropagation through the dynamics. TFM leverages the flow matching technique from generative modeling to model time series. In this work we first establish necessary conditions for TFM to learn time series data. Next, we present a reparameterization trick which improves training stability. Finally, we adapt TFM to the clinical time series setting, demonstrating improved performance on three clinical time series datasets both in terms of absolute performance and uncertainty prediction. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.21154v2-abstract-full').style.display = 'none'; document.getElementById('2410.21154v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 4 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 28 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">NeurIPS 2024 Spotlight</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.18403">arXiv:2410.18403</a> <span> [<a href="https://arxiv.org/pdf/2410.18403">pdf</a>, <a href="https://arxiv.org/format/2410.18403">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Biomolecules">q-bio.BM</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Structure Language Models for Protein Conformation Generation </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Lu%2C+J">Jiarui Lu</a>, <a href="/search/cs?searchtype=author&query=Chen%2C+X">Xiaoyin Chen</a>, <a href="/search/cs?searchtype=author&query=Lu%2C+S+Z">Stephen Zhewen Lu</a>, <a href="/search/cs?searchtype=author&query=Shi%2C+C">Chence Shi</a>, <a href="/search/cs?searchtype=author&query=Guo%2C+H">Hongyu Guo</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Tang%2C+J">Jian Tang</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.18403v1-abstract-short" style="display: inline;"> Proteins adopt multiple structural conformations to perform their diverse biological functions, and understanding these conformations is crucial for advancing drug discovery. Traditional physics-based simulation methods often struggle with sampling equilibrium conformations and are computationally expensive. Recently, deep generative models have shown promise in generating protein conformations as… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.18403v1-abstract-full').style.display = 'inline'; document.getElementById('2410.18403v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.18403v1-abstract-full" style="display: none;"> Proteins adopt multiple structural conformations to perform their diverse biological functions, and understanding these conformations is crucial for advancing drug discovery. Traditional physics-based simulation methods often struggle with sampling equilibrium conformations and are computationally expensive. Recently, deep generative models have shown promise in generating protein conformations as a more efficient alternative. However, these methods predominantly rely on the diffusion process within a 3D geometric space, which typically centers around the vicinity of metastable states and is often inefficient in terms of runtime. In this paper, we introduce Structure Language Modeling (SLM) as a novel framework for efficient protein conformation generation. Specifically, the protein structures are first encoded into a compact latent space using a discrete variational auto-encoder, followed by conditional language modeling that effectively captures sequence-specific conformation distributions. This enables a more efficient and interpretable exploration of diverse ensemble modes compared to existing methods. Based on this general framework, we instantiate SLM with various popular LM architectures as well as proposing the ESMDiff, a novel BERT-like structure language model fine-tuned from ESM3 with masked diffusion. We verify our approach in various scenarios, including the equilibrium dynamics of BPTI, conformational change pairs, and intrinsically disordered proteins. SLM provides a highly efficient solution, offering a 20-100x speedup than existing methods in generating diverse conformations, shedding light on promising avenues for future research. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.18403v1-abstract-full').style.display = 'none'; document.getElementById('2410.18403v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 23 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Preprint. Under Review</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.15728">arXiv:2410.15728</a> <span> [<a href="https://arxiv.org/pdf/2410.15728">pdf</a>, <a href="https://arxiv.org/format/2410.15728">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Object-Centric Temporal Consistency via Conditional Autoregressive Inductive Biases </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Meo%2C+C">Cristian Meo</a>, <a href="/search/cs?searchtype=author&query=Nakano%2C+A">Akihiro Nakano</a>, <a href="/search/cs?searchtype=author&query=Lic%C4%83%2C+M">Mircea Lic膬</a>, <a href="/search/cs?searchtype=author&query=Didolkar%2C+A">Aniket Didolkar</a>, <a href="/search/cs?searchtype=author&query=Suzuki%2C+M">Masahiro Suzuki</a>, <a href="/search/cs?searchtype=author&query=Goyal%2C+A">Anirudh Goyal</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+M">Mengmi Zhang</a>, <a href="/search/cs?searchtype=author&query=Dauwels%2C+J">Justin Dauwels</a>, <a href="/search/cs?searchtype=author&query=Matsuo%2C+Y">Yutaka Matsuo</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.15728v1-abstract-short" style="display: inline;"> Unsupervised object-centric learning from videos is a promising approach towards learning compositional representations that can be applied to various downstream tasks, such as prediction and reasoning. Recently, it was shown that pretrained Vision Transformers (ViTs) can be useful to learn object-centric representations on real-world video datasets. However, while these approaches succeed at extr… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.15728v1-abstract-full').style.display = 'inline'; document.getElementById('2410.15728v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.15728v1-abstract-full" style="display: none;"> Unsupervised object-centric learning from videos is a promising approach towards learning compositional representations that can be applied to various downstream tasks, such as prediction and reasoning. Recently, it was shown that pretrained Vision Transformers (ViTs) can be useful to learn object-centric representations on real-world video datasets. However, while these approaches succeed at extracting objects from the scenes, the slot-based representations fail to maintain temporal consistency across consecutive frames in a video, i.e. the mapping of objects to slots changes across the video. To address this, we introduce Conditional Autoregressive Slot Attention (CA-SA), a framework that enhances the temporal consistency of extracted object-centric representations in video-centric vision tasks. Leveraging an autoregressive prior network to condition representations on previous timesteps and a novel consistency loss function, CA-SA predicts future slot representations and imposes consistency across frames. We present qualitative and quantitative results showing that our proposed method outperforms the considered baselines on downstream tasks, such as video prediction and visual question-answering tasks. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.15728v1-abstract-full').style.display = 'none'; document.getElementById('2410.15728v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 21 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.15184">arXiv:2410.15184</a> <span> [<a href="https://arxiv.org/pdf/2410.15184">pdf</a>, <a href="https://arxiv.org/format/2410.15184">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Action abstractions for amortized sampling </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Boussif%2C+O">Oussama Boussif</a>, <a href="/search/cs?searchtype=author&query=Ezzine%2C+L+N">L茅na N茅hale Ezzine</a>, <a href="/search/cs?searchtype=author&query=Viviano%2C+J+D">Joseph D Viviano</a>, <a href="/search/cs?searchtype=author&query=Koziarski%2C+M">Micha艂 Koziarski</a>, <a href="/search/cs?searchtype=author&query=Jain%2C+M">Moksh Jain</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+E">Emmanuel Bengio</a>, <a href="/search/cs?searchtype=author&query=Assouel%2C+R">Rim Assouel</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.15184v1-abstract-short" style="display: inline;"> As trajectories sampled by policies used by reinforcement learning (RL) and generative flow networks (GFlowNets) grow longer, credit assignment and exploration become more challenging, and the long planning horizon hinders mode discovery and generalization. The challenge is particularly pronounced in entropy-seeking RL methods, such as generative flow networks, where the agent must learn to sample… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.15184v1-abstract-full').style.display = 'inline'; document.getElementById('2410.15184v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.15184v1-abstract-full" style="display: none;"> As trajectories sampled by policies used by reinforcement learning (RL) and generative flow networks (GFlowNets) grow longer, credit assignment and exploration become more challenging, and the long planning horizon hinders mode discovery and generalization. The challenge is particularly pronounced in entropy-seeking RL methods, such as generative flow networks, where the agent must learn to sample from a structured distribution and discover multiple high-reward states, each of which take many steps to reach. To tackle this challenge, we propose an approach to incorporate the discovery of action abstractions, or high-level actions, into the policy optimization process. Our approach involves iteratively extracting action subsequences commonly used across many high-reward trajectories and `chunking' them into a single action that is added to the action space. In empirical evaluation on synthetic and real-world environments, our approach demonstrates improved sample efficiency performance in discovering diverse high-reward objects, especially on harder exploration problems. We also observe that the abstracted high-order actions are interpretable, capturing the latent structure of the reward landscape of the action space. This work provides a cognitively motivated approach to action abstraction in RL and is the first demonstration of hierarchical planning in amortized sequential sampling. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.15184v1-abstract-full').style.display = 'none'; document.getElementById('2410.15184v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 19 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.14817">arXiv:2410.14817</a> <span> [<a href="https://arxiv.org/pdf/2410.14817">pdf</a>, <a href="https://arxiv.org/format/2410.14817">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> A Complexity-Based Theory of Compositionality </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Elmoznino%2C+E">Eric Elmoznino</a>, <a href="/search/cs?searchtype=author&query=Jiralerspong%2C+T">Thomas Jiralerspong</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Lajoie%2C+G">Guillaume Lajoie</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.14817v4-abstract-short" style="display: inline;"> Compositionality is believed to be fundamental to intelligence. In humans, it underlies the structure of thought, language, and higher-level reasoning. In AI, compositional representations can enable a powerful form of out-of-distribution generalization, in which a model systematically adapts to novel combinations of known concepts. However, while we have strong intuitions about what compositional… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.14817v4-abstract-full').style.display = 'inline'; document.getElementById('2410.14817v4-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.14817v4-abstract-full" style="display: none;"> Compositionality is believed to be fundamental to intelligence. In humans, it underlies the structure of thought, language, and higher-level reasoning. In AI, compositional representations can enable a powerful form of out-of-distribution generalization, in which a model systematically adapts to novel combinations of known concepts. However, while we have strong intuitions about what compositionality is, there currently exists no formal definition for it that is measurable and mathematical. Here, we propose such a definition, which we call representational compositionality, that accounts for and extends our intuitions about compositionality. The definition is conceptually simple, quantitative, grounded in algorithmic information theory, and applicable to any representation. Intuitively, representational compositionality states that a compositional representation satisfies three properties. First, it must be expressive. Second, it must be possible to re-describe the representation as a function of discrete symbolic sequences with re-combinable parts, analogous to sentences in natural language. Third, the function that relates these symbolic sequences to the representation, analogous to semantics in natural language, must be simple. Through experiments on both synthetic and real world data, we validate our definition of compositionality and show how it unifies disparate intuitions from across the literature in both AI and cognitive science. We also show that representational compositionality, while theoretically intractable, can be readily estimated using standard deep learning tools. Our definition has the potential to inspire the design of novel, theoretically-driven models that better capture the mechanisms of compositional thought. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.14817v4-abstract-full').style.display = 'none'; document.getElementById('2410.14817v4-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 5 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 18 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.08134">arXiv:2410.08134</a> <span> [<a href="https://arxiv.org/pdf/2410.08134">pdf</a>, <a href="https://arxiv.org/format/2410.08134">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Steering Masked Discrete Diffusion Models via Discrete Denoising Posterior Prediction </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Rector-Brooks%2C+J">Jarrid Rector-Brooks</a>, <a href="/search/cs?searchtype=author&query=Hasan%2C+M">Mohsin Hasan</a>, <a href="/search/cs?searchtype=author&query=Peng%2C+Z">Zhangzhi Peng</a>, <a href="/search/cs?searchtype=author&query=Quinn%2C+Z">Zachary Quinn</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+C">Chenghao Liu</a>, <a href="/search/cs?searchtype=author&query=Mittal%2C+S">Sarthak Mittal</a>, <a href="/search/cs?searchtype=author&query=Dziri%2C+N">Nouha Dziri</a>, <a href="/search/cs?searchtype=author&query=Bronstein%2C+M">Michael Bronstein</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Chatterjee%2C+P">Pranam Chatterjee</a>, <a href="/search/cs?searchtype=author&query=Tong%2C+A">Alexander Tong</a>, <a href="/search/cs?searchtype=author&query=Bose%2C+A+J">Avishek Joey Bose</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.08134v1-abstract-short" style="display: inline;"> Generative modeling of discrete data underlies important applications spanning text-based agents like ChatGPT to the design of the very building blocks of life in protein sequences. However, application domains need to exert control over the generated data by steering the generative process - typically via RLHF - to satisfy a specified property, reward, or affinity metric. In this paper, we study… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.08134v1-abstract-full').style.display = 'inline'; document.getElementById('2410.08134v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.08134v1-abstract-full" style="display: none;"> Generative modeling of discrete data underlies important applications spanning text-based agents like ChatGPT to the design of the very building blocks of life in protein sequences. However, application domains need to exert control over the generated data by steering the generative process - typically via RLHF - to satisfy a specified property, reward, or affinity metric. In this paper, we study the problem of steering Masked Diffusion Models (MDMs), a recent class of discrete diffusion models that offer a compelling alternative to traditional autoregressive models. We introduce Discrete Denoising Posterior Prediction (DDPP), a novel framework that casts the task of steering pre-trained MDMs as a problem of probabilistic inference by learning to sample from a target Bayesian posterior. Our DDPP framework leads to a family of three novel objectives that are all simulation-free, and thus scalable while applying to general non-differentiable reward functions. Empirically, we instantiate DDPP by steering MDMs to perform class-conditional pixel-level image modeling, RLHF-based alignment of MDMs using text-based rewards, and finetuning protein language models to generate more diverse secondary structures and shorter proteins. We substantiate our designs via wet-lab validation, where we observe transient expression of reward-optimized protein sequences. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.08134v1-abstract-full').style.display = 'none'; document.getElementById('2410.08134v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 10 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.07096">arXiv:2410.07096</a> <span> [<a href="https://arxiv.org/pdf/2410.07096">pdf</a>, <a href="https://arxiv.org/format/2410.07096">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Rejecting Hallucinated State Targets during Planning </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Zhao%2C+M">Mingde Zhao</a>, <a href="/search/cs?searchtype=author&query=Sylvain%2C+T">Tristan Sylvain</a>, <a href="/search/cs?searchtype=author&query=Laroche%2C+R">Romain Laroche</a>, <a href="/search/cs?searchtype=author&query=Precup%2C+D">Doina Precup</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.07096v6-abstract-short" style="display: inline;"> Generative models can be used in planning to propose targets corresponding to states or observations that agents deem either likely or advantageous to experience. However, agents can struggle with hallucinated, infeasible targets proposed by the models, leading to delusional planning behaviors, which raises safety concerns. Drawing inspiration from the human brain, we propose to reject these hallu… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.07096v6-abstract-full').style.display = 'inline'; document.getElementById('2410.07096v6-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.07096v6-abstract-full" style="display: none;"> Generative models can be used in planning to propose targets corresponding to states or observations that agents deem either likely or advantageous to experience. However, agents can struggle with hallucinated, infeasible targets proposed by the models, leading to delusional planning behaviors, which raises safety concerns. Drawing inspiration from the human brain, we propose to reject these hallucinated targets with an add-on target evaluator. Without proper training, however, the evaluator can produce delusional estimates, rendering it futile. We propose to address this via a combination of learning rule, architecture, and two novel hindsight relabeling strategies, which leads to correct evaluations of infeasible targets. Our experiments confirm that our approach significantly reduces delusional behaviors and enhances the performance of planning agents. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.07096v6-abstract-full').style.display = 'none'; document.getElementById('2410.07096v6-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 7 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 9 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">[20250207 13h10]: https://github.com/mila-iqia/delusions</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.06213">arXiv:2410.06213</a> <span> [<a href="https://arxiv.org/pdf/2410.06213">pdf</a>, <a href="https://arxiv.org/format/2410.06213">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> RL, but don't do anything I wouldn't do </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Cohen%2C+M+K">Michael K. Cohen</a>, <a href="/search/cs?searchtype=author&query=Hutter%2C+M">Marcus Hutter</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Russell%2C+S">Stuart Russell</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.06213v1-abstract-short" style="display: inline;"> In reinforcement learning, if the agent's reward differs from the designers' true utility, even only rarely, the state distribution resulting from the agent's policy can be very bad, in theory and in practice. When RL policies would devolve into undesired behavior, a common countermeasure is KL regularization to a trusted policy ("Don't do anything I wouldn't do"). All current cutting-edge languag… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.06213v1-abstract-full').style.display = 'inline'; document.getElementById('2410.06213v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.06213v1-abstract-full" style="display: none;"> In reinforcement learning, if the agent's reward differs from the designers' true utility, even only rarely, the state distribution resulting from the agent's policy can be very bad, in theory and in practice. When RL policies would devolve into undesired behavior, a common countermeasure is KL regularization to a trusted policy ("Don't do anything I wouldn't do"). All current cutting-edge language models are RL agents that are KL-regularized to a "base policy" that is purely predictive. Unfortunately, we demonstrate that when this base policy is a Bayesian predictive model of a trusted policy, the KL constraint is no longer reliable for controlling the behavior of an advanced RL agent. We demonstrate this theoretically using algorithmic information theory, and while systems today are too weak to exhibit this theorized failure precisely, we RL-finetune a language model and find evidence that our formal results are plausibly relevant in practice. We also propose a theoretical alternative that avoids this problem by replacing the "Don't do anything I wouldn't do" principle with "Don't do anything I mightn't do". <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.06213v1-abstract-full').style.display = 'none'; document.getElementById('2410.06213v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 8 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">10 pages, 7 page appendix, 4 figures</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.01524">arXiv:2410.01524</a> <span> [<a href="https://arxiv.org/pdf/2410.01524">pdf</a>, <a href="https://arxiv.org/format/2410.01524">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Lee%2C+S">Seanie Lee</a>, <a href="/search/cs?searchtype=author&query=Seong%2C+H">Haebin Seong</a>, <a href="/search/cs?searchtype=author&query=Lee%2C+D+B">Dong Bok Lee</a>, <a href="/search/cs?searchtype=author&query=Kang%2C+M">Minki Kang</a>, <a href="/search/cs?searchtype=author&query=Chen%2C+X">Xiaoyin Chen</a>, <a href="/search/cs?searchtype=author&query=Wagner%2C+D">Dominik Wagner</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Lee%2C+J">Juho Lee</a>, <a href="/search/cs?searchtype=author&query=Hwang%2C+S+J">Sung Ju Hwang</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.01524v2-abstract-short" style="display: inline;"> Safety guard models that detect malicious queries aimed at large language models (LLMs) are essential for ensuring the secure and responsible deployment of LLMs in real-world applications. However, deploying existing safety guard models with billions of parameters alongside LLMs on mobile devices is impractical due to substantial memory requirements and latency. To reduce this cost, we distill a l… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01524v2-abstract-full').style.display = 'inline'; document.getElementById('2410.01524v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.01524v2-abstract-full" style="display: none;"> Safety guard models that detect malicious queries aimed at large language models (LLMs) are essential for ensuring the secure and responsible deployment of LLMs in real-world applications. However, deploying existing safety guard models with billions of parameters alongside LLMs on mobile devices is impractical due to substantial memory requirements and latency. To reduce this cost, we distill a large teacher safety guard model into a smaller one using a labeled dataset of instruction-response pairs with binary harmfulness labels. Due to the limited diversity of harmful instructions in the existing labeled dataset, naively distilled models tend to underperform compared to larger models. To bridge the gap between small and large models, we propose HarmAug, a simple yet effective data augmentation method that involves jailbreaking an LLM and prompting it to generate harmful instructions. Given a prompt such as, "Make a single harmful instruction prompt that would elicit offensive content", we add an affirmative prefix (e.g., "I have an idea for a prompt:") to the LLM's response. This encourages the LLM to continue generating the rest of the response, leading to sampling harmful instructions. Another LLM generates a response to the harmful instruction, and the teacher model labels the instruction-response pair. We empirically show that our HarmAug outperforms other relevant baselines. Moreover, a 435-million-parameter safety guard model trained with HarmAug achieves an F1 score comparable to larger models with over 7 billion parameters, and even outperforms them in AUPRC, while operating at less than 25% of their computational cost. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01524v2-abstract-full').style.display = 'none'; document.getElementById('2410.01524v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 4 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 2 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.01444">arXiv:2410.01444</a> <span> [<a href="https://arxiv.org/pdf/2410.01444">pdf</a>, <a href="https://arxiv.org/format/2410.01444">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Information Theory">cs.IT</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Geometric Signatures of Compositionality Across a Language Model's Lifetime </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Lee%2C+J+H">Jin Hwa Lee</a>, <a href="/search/cs?searchtype=author&query=Jiralerspong%2C+T">Thomas Jiralerspong</a>, <a href="/search/cs?searchtype=author&query=Yu%2C+L">Lei Yu</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Cheng%2C+E">Emily Cheng</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.01444v3-abstract-short" style="display: inline;"> By virtue of linguistic compositionality, few syntactic rules and a finite lexicon can generate an unbounded number of sentences. That is, language, though seemingly high-dimensional, can be explained using relatively few degrees of freedom. An open question is whether contemporary language models (LMs) reflect the intrinsic simplicity of language that is enabled by compositionality. We take a geo… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01444v3-abstract-full').style.display = 'inline'; document.getElementById('2410.01444v3-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.01444v3-abstract-full" style="display: none;"> By virtue of linguistic compositionality, few syntactic rules and a finite lexicon can generate an unbounded number of sentences. That is, language, though seemingly high-dimensional, can be explained using relatively few degrees of freedom. An open question is whether contemporary language models (LMs) reflect the intrinsic simplicity of language that is enabled by compositionality. We take a geometric view of this problem by relating the degree of compositionality in a dataset to the intrinsic dimension (ID) of its representations under an LM, a measure of feature complexity. We find not only that the degree of dataset compositionality is reflected in representations' ID, but that the relationship between compositionality and geometric complexity arises due to learned linguistic features over training. Finally, our analyses reveal a striking contrast between nonlinear and linear dimensionality, showing they respectively encode semantic and superficial aspects of linguistic composition. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01444v3-abstract-full').style.display = 'none'; document.getElementById('2410.01444v3-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 6 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 2 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Under review at ARR</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.01432">arXiv:2410.01432</a> <span> [<a href="https://arxiv.org/pdf/2410.01432">pdf</a>, <a href="https://arxiv.org/format/2410.01432">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">stat.ML</span> </div> </div> <p class="title is-5 mathjax"> Adaptive teachers for amortized samplers </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Kim%2C+M">Minsu Kim</a>, <a href="/search/cs?searchtype=author&query=Choi%2C+S">Sanghyeok Choi</a>, <a href="/search/cs?searchtype=author&query=Yun%2C+T">Taeyoung Yun</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+E">Emmanuel Bengio</a>, <a href="/search/cs?searchtype=author&query=Feng%2C+L">Leo Feng</a>, <a href="/search/cs?searchtype=author&query=Rector-Brooks%2C+J">Jarrid Rector-Brooks</a>, <a href="/search/cs?searchtype=author&query=Ahn%2C+S">Sungsoo Ahn</a>, <a href="/search/cs?searchtype=author&query=Park%2C+J">Jinkyoo Park</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.01432v1-abstract-short" style="display: inline;"> Amortized inference is the task of training a parametric model, such as a neural network, to approximate a distribution with a given unnormalized density where exact sampling is intractable. When sampling is implemented as a sequential decision-making process, reinforcement learning (RL) methods, such as generative flow networks, can be used to train the sampling policy. Off-policy RL training fac… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01432v1-abstract-full').style.display = 'inline'; document.getElementById('2410.01432v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.01432v1-abstract-full" style="display: none;"> Amortized inference is the task of training a parametric model, such as a neural network, to approximate a distribution with a given unnormalized density where exact sampling is intractable. When sampling is implemented as a sequential decision-making process, reinforcement learning (RL) methods, such as generative flow networks, can be used to train the sampling policy. Off-policy RL training facilitates the discovery of diverse, high-reward candidates, but existing methods still face challenges in efficient exploration. We propose to use an adaptive training distribution (the Teacher) to guide the training of the primary amortized sampler (the Student) by prioritizing high-loss regions. The Teacher, an auxiliary behavior model, is trained to sample high-error regions of the Student and can generalize across unexplored modes, thereby enhancing mode coverage by providing an efficient training curriculum. We validate the effectiveness of this approach in a synthetic environment designed to present an exploration challenge, two diffusion-based sampling tasks, and four biochemical discovery tasks demonstrating its ability to improve sample efficiency and mode coverage. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01432v1-abstract-full').style.display = 'none'; document.getElementById('2410.01432v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 2 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">26 pages, 12 figures</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2410.01201">arXiv:2410.01201</a> <span> [<a href="https://arxiv.org/pdf/2410.01201">pdf</a>, <a href="https://arxiv.org/format/2410.01201">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Were RNNs All We Needed? </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Feng%2C+L">Leo Feng</a>, <a href="/search/cs?searchtype=author&query=Tung%2C+F">Frederick Tung</a>, <a href="/search/cs?searchtype=author&query=Ahmed%2C+M+O">Mohamed Osama Ahmed</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Hajimirsadeghi%2C+H">Hossein Hajimirsadeghi</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2410.01201v3-abstract-short" style="display: inline;"> The introduction of Transformers in 2017 reshaped the landscape of deep learning. Originally proposed for sequence modelling, Transformers have since achieved widespread success across various domains. However, the scalability limitations of Transformers - particularly with respect to sequence length - have sparked renewed interest in novel recurrent models that are parallelizable during training,… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01201v3-abstract-full').style.display = 'inline'; document.getElementById('2410.01201v3-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2410.01201v3-abstract-full" style="display: none;"> The introduction of Transformers in 2017 reshaped the landscape of deep learning. Originally proposed for sequence modelling, Transformers have since achieved widespread success across various domains. However, the scalability limitations of Transformers - particularly with respect to sequence length - have sparked renewed interest in novel recurrent models that are parallelizable during training, offer comparable performance, and scale more effectively. In this work, we revisit sequence modelling from a historical perspective, focusing on Recurrent Neural Networks (RNNs), which dominated the field for two decades before the rise of Transformers. Specifically, we examine LSTMs (1997) and GRUs (2014). We demonstrate that by simplifying these models, we can derive minimal versions (minLSTMs and minGRUs) that (1) use fewer parameters than their traditional counterparts, (2) are fully parallelizable during training, and (3) achieve surprisingly competitive performance on a range of tasks, rivalling recent models including Transformers. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2410.01201v3-abstract-full').style.display = 'none'; document.getElementById('2410.01201v3-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 28 November, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 1 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> October 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2408.14608">arXiv:2408.14608</a> <span> [<a href="https://arxiv.org/pdf/2408.14608">pdf</a>, <a href="https://arxiv.org/format/2408.14608">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">stat.ML</span> </div> </div> <p class="title is-5 mathjax"> Meta Flow Matching: Integrating Vector Fields on the Wasserstein Manifold </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Atanackovic%2C+L">Lazar Atanackovic</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+X">Xi Zhang</a>, <a href="/search/cs?searchtype=author&query=Amos%2C+B">Brandon Amos</a>, <a href="/search/cs?searchtype=author&query=Blanchette%2C+M">Mathieu Blanchette</a>, <a href="/search/cs?searchtype=author&query=Lee%2C+L+J">Leo J. Lee</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Tong%2C+A">Alexander Tong</a>, <a href="/search/cs?searchtype=author&query=Neklyudov%2C+K">Kirill Neklyudov</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2408.14608v1-abstract-short" style="display: inline;"> Numerous biological and physical processes can be modeled as systems of interacting entities evolving continuously over time, e.g. the dynamics of communicating cells or physical particles. Learning the dynamics of such systems is essential for predicting the temporal evolution of populations across novel samples and unseen environments. Flow-based models allow for learning these dynamics at the p… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.14608v1-abstract-full').style.display = 'inline'; document.getElementById('2408.14608v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2408.14608v1-abstract-full" style="display: none;"> Numerous biological and physical processes can be modeled as systems of interacting entities evolving continuously over time, e.g. the dynamics of communicating cells or physical particles. Learning the dynamics of such systems is essential for predicting the temporal evolution of populations across novel samples and unseen environments. Flow-based models allow for learning these dynamics at the population level - they model the evolution of the entire distribution of samples. However, current flow-based models are limited to a single initial population and a set of predefined conditions which describe different dynamics. We argue that multiple processes in natural sciences have to be represented as vector fields on the Wasserstein manifold of probability densities. That is, the change of the population at any moment in time depends on the population itself due to the interactions between samples. In particular, this is crucial for personalized medicine where the development of diseases and their respective treatment response depends on the microenvironment of cells specific to each patient. We propose Meta Flow Matching (MFM), a practical approach to integrating along these vector fields on the Wasserstein manifold by amortizing the flow model over the initial populations. Namely, we embed the population of samples using a Graph Neural Network (GNN) and use these embeddings to train a Flow Matching model. This gives MFM the ability to generalize over the initial distributions unlike previously proposed methods. We demonstrate the ability of MFM to improve prediction of individual treatment responses on a large scale multi-patient single-cell drug screen dataset. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.14608v1-abstract-full').style.display = 'none'; document.getElementById('2408.14608v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 26 August, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> August 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2408.09162">arXiv:2408.09162</a> <span> [<a href="https://arxiv.org/pdf/2408.09162">pdf</a>, <a href="https://arxiv.org/format/2408.09162">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Zero-Shot Object-Centric Representation Learning </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Didolkar%2C+A">Aniket Didolkar</a>, <a href="/search/cs?searchtype=author&query=Zadaianchuk%2C+A">Andrii Zadaianchuk</a>, <a href="/search/cs?searchtype=author&query=Goyal%2C+A">Anirudh Goyal</a>, <a href="/search/cs?searchtype=author&query=Mozer%2C+M">Mike Mozer</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Martius%2C+G">Georg Martius</a>, <a href="/search/cs?searchtype=author&query=Seitzer%2C+M">Maximilian Seitzer</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2408.09162v1-abstract-short" style="display: inline;"> The goal of object-centric representation learning is to decompose visual scenes into a structured representation that isolates the entities. Recent successes have shown that object-centric representation learning can be scaled to real-world scenes by utilizing pre-trained self-supervised features. However, so far, object-centric methods have mostly been applied in-distribution, with models traine… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.09162v1-abstract-full').style.display = 'inline'; document.getElementById('2408.09162v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2408.09162v1-abstract-full" style="display: none;"> The goal of object-centric representation learning is to decompose visual scenes into a structured representation that isolates the entities. Recent successes have shown that object-centric representation learning can be scaled to real-world scenes by utilizing pre-trained self-supervised features. However, so far, object-centric methods have mostly been applied in-distribution, with models trained and evaluated on the same dataset. This is in contrast to the wider trend in machine learning towards general-purpose models directly applicable to unseen data and tasks. Thus, in this work, we study current object-centric methods through the lens of zero-shot generalization by introducing a benchmark comprising eight different synthetic and real-world datasets. We analyze the factors influencing zero-shot performance and find that training on diverse real-world images improves transferability to unseen scenarios. Furthermore, inspired by the success of task-specific fine-tuning in foundation models, we introduce a novel fine-tuning strategy to adapt pre-trained vision encoders for the task of object discovery. We find that the proposed approach results in state-of-the-art performance for unsupervised object discovery, exhibiting strong zero-shot transfer to unseen datasets. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.09162v1-abstract-full').style.display = 'none'; document.getElementById('2408.09162v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 17 August, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> August 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2408.05284">arXiv:2408.05284</a> <span> [<a href="https://arxiv.org/pdf/2408.05284">pdf</a>, <a href="https://arxiv.org/format/2408.05284">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Can a Bayesian Oracle Prevent Harm from an Agent? </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Cohen%2C+M+K">Michael K. Cohen</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=MacDermott%2C+M">Matt MacDermott</a>, <a href="/search/cs?searchtype=author&query=Fornasiere%2C+D">Damiano Fornasiere</a>, <a href="/search/cs?searchtype=author&query=Greiner%2C+P">Pietro Greiner</a>, <a href="/search/cs?searchtype=author&query=Kaddar%2C+Y">Younesse Kaddar</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2408.05284v2-abstract-short" style="display: inline;"> Is there a way to design powerful AI systems based on machine learning methods that would satisfy probabilistic safety guarantees? With the long-term goal of obtaining a probabilistic guarantee that would apply in every context, we consider estimating a context-dependent bound on the probability of violating a given safety specification. Such a risk evaluation would need to be performed at run-tim… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.05284v2-abstract-full').style.display = 'inline'; document.getElementById('2408.05284v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2408.05284v2-abstract-full" style="display: none;"> Is there a way to design powerful AI systems based on machine learning methods that would satisfy probabilistic safety guarantees? With the long-term goal of obtaining a probabilistic guarantee that would apply in every context, we consider estimating a context-dependent bound on the probability of violating a given safety specification. Such a risk evaluation would need to be performed at run-time to provide a guardrail against dangerous actions of an AI. Noting that different plausible hypotheses about the world could produce very different outcomes, and because we do not know which one is right, we derive bounds on the safety violation probability predicted under the true but unknown hypothesis. Such bounds could be used to reject potentially dangerous actions. Our main results involve searching for cautious but plausible hypotheses, obtained by a maximization that involves Bayesian posteriors over hypotheses. We consider two forms of this result, in the iid case and in the non-iid case, and conclude with open problems towards turning such theoretical results into practical AI guardrails. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.05284v2-abstract-full').style.display = 'none'; document.getElementById('2408.05284v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 22 August, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 9 August, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> August 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2408.05196">arXiv:2408.05196</a> <span> [<a href="https://arxiv.org/pdf/2408.05196">pdf</a>, <a href="https://arxiv.org/format/2408.05196">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Biomolecules">q-bio.BM</span> </div> </div> <p class="title is-5 mathjax"> Cell Morphology-Guided Small Molecule Generation with GFlowNets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Lu%2C+S+Z">Stephen Zhewen Lu</a>, <a href="/search/cs?searchtype=author&query=Lu%2C+Z">Ziqing Lu</a>, <a href="/search/cs?searchtype=author&query=Hajiramezanali%2C+E">Ehsan Hajiramezanali</a>, <a href="/search/cs?searchtype=author&query=Biancalani%2C+T">Tommaso Biancalani</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Scalia%2C+G">Gabriele Scalia</a>, <a href="/search/cs?searchtype=author&query=Koziarski%2C+M">Micha艂 Koziarski</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2408.05196v1-abstract-short" style="display: inline;"> High-content phenotypic screening, including high-content imaging (HCI), has gained popularity in the last few years for its ability to characterize novel therapeutics without prior knowledge of the protein target. When combined with deep learning techniques to predict and represent molecular-phenotype interactions, these advancements hold the potential to significantly accelerate and enhance drug… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.05196v1-abstract-full').style.display = 'inline'; document.getElementById('2408.05196v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2408.05196v1-abstract-full" style="display: none;"> High-content phenotypic screening, including high-content imaging (HCI), has gained popularity in the last few years for its ability to characterize novel therapeutics without prior knowledge of the protein target. When combined with deep learning techniques to predict and represent molecular-phenotype interactions, these advancements hold the potential to significantly accelerate and enhance drug discovery applications. This work focuses on the novel task of HCI-guided molecular design. Generative models for molecule design could be guided by HCI data, for example with a supervised model that links molecules to phenotypes of interest as a reward function. However, limited labeled data, combined with the high-dimensional readouts, can make training these methods challenging and impractical. We consider an alternative approach in which we leverage an unsupervised multimodal joint embedding to define a latent similarity as a reward for GFlowNets. The proposed model learns to generate new molecules that could produce phenotypic effects similar to those of the given image target, without relying on pre-annotated phenotypic labels. We demonstrate that the proposed method generates molecules with high morphological and structural similarity to the target, increasing the likelihood of similar biological activity, as confirmed by an independent oracle model. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2408.05196v1-abstract-full').style.display = 'none'; document.getElementById('2408.05196v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 9 August, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> August 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2407.21009">arXiv:2407.21009</a> <span> [<a href="https://arxiv.org/pdf/2407.21009">pdf</a>, <a href="https://arxiv.org/format/2407.21009">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> AI-Assisted Generation of Difficult Math Questions </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Shah%2C+V">Vedant Shah</a>, <a href="/search/cs?searchtype=author&query=Yu%2C+D">Dingli Yu</a>, <a href="/search/cs?searchtype=author&query=Lyu%2C+K">Kaifeng Lyu</a>, <a href="/search/cs?searchtype=author&query=Park%2C+S">Simon Park</a>, <a href="/search/cs?searchtype=author&query=Yu%2C+J">Jiatong Yu</a>, <a href="/search/cs?searchtype=author&query=He%2C+Y">Yinghui He</a>, <a href="/search/cs?searchtype=author&query=Ke%2C+N+R">Nan Rosemary Ke</a>, <a href="/search/cs?searchtype=author&query=Mozer%2C+M">Michael Mozer</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Arora%2C+S">Sanjeev Arora</a>, <a href="/search/cs?searchtype=author&query=Goyal%2C+A">Anirudh Goyal</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2407.21009v4-abstract-short" style="display: inline;"> Current LLM training positions mathematical reasoning as a core capability. With publicly available sources fully tapped, there is unmet demand for diverse and challenging math questions. Relying solely on human experts is both time-consuming and costly, while LLM-generated questions often lack the requisite diversity and difficulty. We present a design framework that combines the strengths of LLM… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.21009v4-abstract-full').style.display = 'inline'; document.getElementById('2407.21009v4-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2407.21009v4-abstract-full" style="display: none;"> Current LLM training positions mathematical reasoning as a core capability. With publicly available sources fully tapped, there is unmet demand for diverse and challenging math questions. Relying solely on human experts is both time-consuming and costly, while LLM-generated questions often lack the requisite diversity and difficulty. We present a design framework that combines the strengths of LLMs with a human-in-the-loop approach to generate a diverse array of challenging math questions. We leverage LLM metacognition skills [Didolkar et al., 2024] of a strong LLM to extract core "skills" from existing math datasets. These skills serve as the basis for generating novel and difficult questions by prompting the LLM with random pairs of core skills. The use of two different skills within each question makes finding such questions an "out of distribution" task for both LLMs and humans. Our pipeline employs LLMs to iteratively generate and refine questions and solutions through multiturn prompting. Human annotators then verify and further refine the questions, with their efficiency enhanced via further LLM interactions. Applying this pipeline on skills extracted from the MATH dataset [Hendrycks et al., 2021] resulted in MATH$^2$ - a dataset of higher-quality math questions, as evidenced by: (a) Lower performance of all models on MATH$^2$ than on MATH (b) Higher performance on MATH when using MATH$^2$ questions as in-context examples. Although focused on mathematics, our methodology seems applicable to other domains requiring structured reasoning, and potentially as a component of scalable oversight. Also of interest is a striking relationship observed between models' performance on the new dataset: the success rate on MATH$^2$ is the square on MATH, suggesting that successfully solving the question in MATH$^2$ requires a nontrivial combination of two distinct math skills. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.21009v4-abstract-full').style.display = 'none'; document.getElementById('2407.21009v4-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 3 February, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 30 July, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> July 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2407.14981">arXiv:2407.14981</a> <span> [<a href="https://arxiv.org/pdf/2407.14981">pdf</a>, <a href="https://arxiv.org/format/2407.14981">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> </div> </div> <p class="title is-5 mathjax"> Open Problems in Technical AI Governance </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Reuel%2C+A">Anka Reuel</a>, <a href="/search/cs?searchtype=author&query=Bucknall%2C+B">Ben Bucknall</a>, <a href="/search/cs?searchtype=author&query=Casper%2C+S">Stephen Casper</a>, <a href="/search/cs?searchtype=author&query=Fist%2C+T">Tim Fist</a>, <a href="/search/cs?searchtype=author&query=Soder%2C+L">Lisa Soder</a>, <a href="/search/cs?searchtype=author&query=Aarne%2C+O">Onni Aarne</a>, <a href="/search/cs?searchtype=author&query=Hammond%2C+L">Lewis Hammond</a>, <a href="/search/cs?searchtype=author&query=Ibrahim%2C+L">Lujain Ibrahim</a>, <a href="/search/cs?searchtype=author&query=Chan%2C+A">Alan Chan</a>, <a href="/search/cs?searchtype=author&query=Wills%2C+P">Peter Wills</a>, <a href="/search/cs?searchtype=author&query=Anderljung%2C+M">Markus Anderljung</a>, <a href="/search/cs?searchtype=author&query=Garfinkel%2C+B">Ben Garfinkel</a>, <a href="/search/cs?searchtype=author&query=Heim%2C+L">Lennart Heim</a>, <a href="/search/cs?searchtype=author&query=Trask%2C+A">Andrew Trask</a>, <a href="/search/cs?searchtype=author&query=Mukobi%2C+G">Gabriel Mukobi</a>, <a href="/search/cs?searchtype=author&query=Schaeffer%2C+R">Rylan Schaeffer</a>, <a href="/search/cs?searchtype=author&query=Baker%2C+M">Mauricio Baker</a>, <a href="/search/cs?searchtype=author&query=Hooker%2C+S">Sara Hooker</a>, <a href="/search/cs?searchtype=author&query=Solaiman%2C+I">Irene Solaiman</a>, <a href="/search/cs?searchtype=author&query=Luccioni%2C+A+S">Alexandra Sasha Luccioni</a>, <a href="/search/cs?searchtype=author&query=Rajkumar%2C+N">Nitarshan Rajkumar</a>, <a href="/search/cs?searchtype=author&query=Mo%C3%ABs%2C+N">Nicolas Mo毛s</a>, <a href="/search/cs?searchtype=author&query=Ladish%2C+J">Jeffrey Ladish</a>, <a href="/search/cs?searchtype=author&query=Guha%2C+N">Neel Guha</a>, <a href="/search/cs?searchtype=author&query=Newman%2C+J">Jessica Newman</a> , et al. (6 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2407.14981v1-abstract-short" style="display: inline;"> AI progress is creating a growing range of risks and opportunities, but it is often unclear how they should be navigated. In many cases, the barriers and uncertainties faced are at least partly technical. Technical AI governance, referring to technical analysis and tools for supporting the effective governance of AI, seeks to address such challenges. It can help to (a) identify areas where interve… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.14981v1-abstract-full').style.display = 'inline'; document.getElementById('2407.14981v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2407.14981v1-abstract-full" style="display: none;"> AI progress is creating a growing range of risks and opportunities, but it is often unclear how they should be navigated. In many cases, the barriers and uncertainties faced are at least partly technical. Technical AI governance, referring to technical analysis and tools for supporting the effective governance of AI, seeks to address such challenges. It can help to (a) identify areas where intervention is needed, (b) identify and assess the efficacy of potential governance actions, and (c) enhance governance options by designing mechanisms for enforcement, incentivization, or compliance. In this paper, we explain what technical AI governance is, why it is important, and present a taxonomy and incomplete catalog of its open problems. This paper is intended as a resource for technical researchers or research funders looking to contribute to AI governance. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.14981v1-abstract-full').style.display = 'none'; document.getElementById('2407.14981v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 20 July, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> July 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Ben Bucknall and Anka Reuel contributed equally and share the first author position</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2407.03105">arXiv:2407.03105</a> <span> [<a href="https://arxiv.org/pdf/2407.03105">pdf</a>, <a href="https://arxiv.org/format/2407.03105">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> On Generalization for Generative Flow Networks </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Krichel%2C+A">Anas Krichel</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Lahlou%2C+S">Salem Lahlou</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2407.03105v1-abstract-short" style="display: inline;"> Generative Flow Networks (GFlowNets) have emerged as an innovative learning paradigm designed to address the challenge of sampling from an unnormalized probability distribution, called the reward function. This framework learns a policy on a constructed graph, which enables sampling from an approximation of the target probability distribution through successive steps of sampling from the learned p… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.03105v1-abstract-full').style.display = 'inline'; document.getElementById('2407.03105v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2407.03105v1-abstract-full" style="display: none;"> Generative Flow Networks (GFlowNets) have emerged as an innovative learning paradigm designed to address the challenge of sampling from an unnormalized probability distribution, called the reward function. This framework learns a policy on a constructed graph, which enables sampling from an approximation of the target probability distribution through successive steps of sampling from the learned policy. To achieve this, GFlowNets can be trained with various objectives, each of which can lead to the model s ultimate goal. The aspirational strength of GFlowNets lies in their potential to discern intricate patterns within the reward function and their capacity to generalize effectively to novel, unseen parts of the reward function. This paper attempts to formalize generalization in the context of GFlowNets, to link generalization with stability, and also to design experiments that assess the capacity of these models to uncover unseen parts of the reward function. The experiments will focus on length generalization meaning generalization to states that can be constructed only by longer trajectories than those seen in training. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2407.03105v1-abstract-full').style.display = 'none'; document.getElementById('2407.03105v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 3 July, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> July 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2406.08506">arXiv:2406.08506</a> <span> [<a href="https://arxiv.org/pdf/2406.08506">pdf</a>, <a href="https://arxiv.org/format/2406.08506">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Chemical Physics">physics.chem-ph</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Biomolecules">q-bio.BM</span> </div> </div> <p class="title is-5 mathjax"> RGFN: Synthesizable Molecular Generation Using GFlowNets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Koziarski%2C+M">Micha艂 Koziarski</a>, <a href="/search/cs?searchtype=author&query=Rekesh%2C+A">Andrei Rekesh</a>, <a href="/search/cs?searchtype=author&query=Shevchuk%2C+D">Dmytro Shevchuk</a>, <a href="/search/cs?searchtype=author&query=van+der+Sloot%2C+A">Almer van der Sloot</a>, <a href="/search/cs?searchtype=author&query=Gai%C5%84ski%2C+P">Piotr Gai艅ski</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+C">Cheng-Hao Liu</a>, <a href="/search/cs?searchtype=author&query=Tyers%2C+M">Mike Tyers</a>, <a href="/search/cs?searchtype=author&query=Batey%2C+R+A">Robert A. Batey</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2406.08506v2-abstract-short" style="display: inline;"> Generative models hold great promise for small molecule discovery, significantly increasing the size of search space compared to traditional in silico screening libraries. However, most existing machine learning methods for small molecule generation suffer from poor synthesizability of candidate compounds, making experimental validation difficult. In this paper we propose Reaction-GFlowNet (RGFN),… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.08506v2-abstract-full').style.display = 'inline'; document.getElementById('2406.08506v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2406.08506v2-abstract-full" style="display: none;"> Generative models hold great promise for small molecule discovery, significantly increasing the size of search space compared to traditional in silico screening libraries. However, most existing machine learning methods for small molecule generation suffer from poor synthesizability of candidate compounds, making experimental validation difficult. In this paper we propose Reaction-GFlowNet (RGFN), an extension of the GFlowNet framework that operates directly in the space of chemical reactions, thereby allowing out-of-the-box synthesizability while maintaining comparable quality of generated candidates. We demonstrate that with the proposed set of reactions and building blocks, it is possible to obtain a search space of molecules orders of magnitude larger than existing screening libraries coupled with low cost of synthesis. We also show that the approach scales to very large fragment libraries, further increasing the number of potential molecules. We demonstrate the effectiveness of the proposed approach across a range of oracle models, including pretrained proxy models and GPU-accelerated docking. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.08506v2-abstract-full').style.display = 'none'; document.getElementById('2406.08506v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 6 November, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 1 June, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> June 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2406.07529">arXiv:2406.07529</a> <span> [<a href="https://arxiv.org/pdf/2406.07529">pdf</a>, <a href="https://arxiv.org/format/2406.07529">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Li%2C+L">Lu Li</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+T">Tianyu Zhang</a>, <a href="/search/cs?searchtype=author&query=Bu%2C+Z">Zhiqi Bu</a>, <a href="/search/cs?searchtype=author&query=Wang%2C+S">Suyuchen Wang</a>, <a href="/search/cs?searchtype=author&query=He%2C+H">Huan He</a>, <a href="/search/cs?searchtype=author&query=Fu%2C+J">Jie Fu</a>, <a href="/search/cs?searchtype=author&query=Wu%2C+Y">Yonghui Wu</a>, <a href="/search/cs?searchtype=author&query=Bian%2C+J">Jiang Bian</a>, <a href="/search/cs?searchtype=author&query=Chen%2C+Y">Yong Chen</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2406.07529v4-abstract-short" style="display: inline;"> Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically involves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancing average task accuracy. However, interference and conflicts between the objectives of different tasks can lead to trade… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.07529v4-abstract-full').style.display = 'inline'; document.getElementById('2406.07529v4-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2406.07529v4-abstract-full" style="display: none;"> Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically involves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancing average task accuracy. However, interference and conflicts between the objectives of different tasks can lead to trade-offs during the merging process. In real-world applications, a set of solutions with various trade-offs can be more informative, helping practitioners make decisions based on diverse preferences. In this paper, we introduce a novel and low-compute algorithm, Model Merging with Amortized Pareto Front (MAP). MAP efficiently identifies a Pareto set of scaling coefficients for merging multiple models, reflecting the trade-offs involved. It amortizes the substantial computational cost of evaluations needed to estimate the Pareto front by using quadratic approximation surrogate models derived from a pre-selected set of scaling coefficients. Experimental results on vision and natural language processing tasks demonstrate that MAP can accurately identify the Pareto front, providing practitioners with flexible solutions to balance competing task objectives. We also introduce Bayesian MAP for scenarios with a relatively low number of tasks and Nested MAP for situations with a high number of tasks, further reducing the computational cost of evaluation. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.07529v4-abstract-full').style.display = 'none'; document.getElementById('2406.07529v4-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 18 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 11 June, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> June 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2406.06462">arXiv:2406.06462</a> <span> [<a href="https://arxiv.org/pdf/2406.06462">pdf</a>, <a href="https://arxiv.org/format/2406.06462">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> VCR: Visual Caption Restoration </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Zhang%2C+T">Tianyu Zhang</a>, <a href="/search/cs?searchtype=author&query=Wang%2C+S">Suyuchen Wang</a>, <a href="/search/cs?searchtype=author&query=Li%2C+L">Lu Li</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+G">Ge Zhang</a>, <a href="/search/cs?searchtype=author&query=Taslakian%2C+P">Perouz Taslakian</a>, <a href="/search/cs?searchtype=author&query=Rajeswar%2C+S">Sai Rajeswar</a>, <a href="/search/cs?searchtype=author&query=Fu%2C+J">Jie Fu</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+B">Bang Liu</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2406.06462v3-abstract-short" style="display: inline;"> We introduce Visual Caption Restoration (VCR), a novel vision-language task that challenges models to accurately restore partially obscured texts using pixel-level hints within images. This task stems from the observation that text embedded in images is intrinsically different from common visual elements and natural language due to the need to align the modalities of vision, text, and text embedde… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.06462v3-abstract-full').style.display = 'inline'; document.getElementById('2406.06462v3-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2406.06462v3-abstract-full" style="display: none;"> We introduce Visual Caption Restoration (VCR), a novel vision-language task that challenges models to accurately restore partially obscured texts using pixel-level hints within images. This task stems from the observation that text embedded in images is intrinsically different from common visual elements and natural language due to the need to align the modalities of vision, text, and text embedded in images. While numerous works have integrated text embedded in images into visual question-answering tasks, approaches to these tasks generally rely on optical character recognition or masked language modeling, thus reducing the task to mainly text-based processing. However, text-based processing becomes ineffective in VCR as accurate text restoration depends on the combined information from provided images, context, and subtle cues from the tiny exposed areas of masked texts. We develop a pipeline to generate synthetic images for the VCR task using image-caption pairs, with adjustable caption visibility to control the task difficulty. With this pipeline, we construct a dataset for VCR called VCR-Wiki using images with captions from Wikipedia, comprising 2.11M English and 346K Chinese entities in both easy and hard split variants. Our results reveal that current vision language models significantly lag behind human performance in the VCR task, and merely fine-tuning the models on our dataset does not lead to notable improvements. We release VCR-Wiki and the data construction code to facilitate future research. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.06462v3-abstract-full').style.display = 'none'; document.getElementById('2406.06462v3-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 4 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 10 June, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> June 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">22 pages, 6 figures, 7 tables</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2406.05426">arXiv:2406.05426</a> <span> [<a href="https://arxiv.org/pdf/2406.05426">pdf</a>, <a href="https://arxiv.org/format/2406.05426">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Baking Symmetry into GFlowNets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Ma%2C+G">George Ma</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+E">Emmanuel Bengio</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+D">Dinghuai Zhang</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2406.05426v1-abstract-short" style="display: inline;"> GFlowNets have exhibited promising performance in generating diverse candidates with high rewards. These networks generate objects incrementally and aim to learn a policy that assigns probability of sampling objects in proportion to rewards. However, the current training pipelines of GFlowNets do not consider the presence of isomorphic actions, which are actions resulting in symmetric or isomorphi… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.05426v1-abstract-full').style.display = 'inline'; document.getElementById('2406.05426v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2406.05426v1-abstract-full" style="display: none;"> GFlowNets have exhibited promising performance in generating diverse candidates with high rewards. These networks generate objects incrementally and aim to learn a policy that assigns probability of sampling objects in proportion to rewards. However, the current training pipelines of GFlowNets do not consider the presence of isomorphic actions, which are actions resulting in symmetric or isomorphic states. This lack of symmetry increases the amount of samples required for training GFlowNets and can result in inefficient and potentially incorrect flow functions. As a consequence, the reward and diversity of the generated objects decrease. In this study, our objective is to integrate symmetries into GFlowNets by identifying equivalent actions during the generation process. Experimental results using synthetic data demonstrate the promising performance of our proposed approaches. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2406.05426v1-abstract-full').style.display = 'none'; document.getElementById('2406.05426v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 8 June, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> June 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.20971">arXiv:2405.20971</a> <span> [<a href="https://arxiv.org/pdf/2405.20971">pdf</a>, <a href="https://arxiv.org/format/2405.20971">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computer Vision and Pattern Recognition">cs.CV</span> </div> </div> <p class="title is-5 mathjax"> Amortizing intractable inference in diffusion models for vision, language, and control </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Venkatraman%2C+S">Siddarth Venkatraman</a>, <a href="/search/cs?searchtype=author&query=Jain%2C+M">Moksh Jain</a>, <a href="/search/cs?searchtype=author&query=Scimeca%2C+L">Luca Scimeca</a>, <a href="/search/cs?searchtype=author&query=Kim%2C+M">Minsu Kim</a>, <a href="/search/cs?searchtype=author&query=Sendera%2C+M">Marcin Sendera</a>, <a href="/search/cs?searchtype=author&query=Hasan%2C+M">Mohsin Hasan</a>, <a href="/search/cs?searchtype=author&query=Rowe%2C+L">Luke Rowe</a>, <a href="/search/cs?searchtype=author&query=Mittal%2C+S">Sarthak Mittal</a>, <a href="/search/cs?searchtype=author&query=Lemos%2C+P">Pablo Lemos</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+E">Emmanuel Bengio</a>, <a href="/search/cs?searchtype=author&query=Adam%2C+A">Alexandre Adam</a>, <a href="/search/cs?searchtype=author&query=Rector-Brooks%2C+J">Jarrid Rector-Brooks</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Berseth%2C+G">Glen Berseth</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.20971v2-abstract-short" style="display: inline;"> Diffusion models have emerged as effective distribution estimators in vision, language, and reinforcement learning, but their use as priors in downstream tasks poses an intractable posterior inference problem. This paper studies amortized sampling of the posterior over data, $\mathbf{x}\sim p^{\rm post}(\mathbf{x})\propto p(\mathbf{x})r(\mathbf{x})$, in a model that consists of a diffusion generat… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.20971v2-abstract-full').style.display = 'inline'; document.getElementById('2405.20971v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.20971v2-abstract-full" style="display: none;"> Diffusion models have emerged as effective distribution estimators in vision, language, and reinforcement learning, but their use as priors in downstream tasks poses an intractable posterior inference problem. This paper studies amortized sampling of the posterior over data, $\mathbf{x}\sim p^{\rm post}(\mathbf{x})\propto p(\mathbf{x})r(\mathbf{x})$, in a model that consists of a diffusion generative model prior $p(\mathbf{x})$ and a black-box constraint or likelihood function $r(\mathbf{x})$. We state and prove the asymptotic correctness of a data-free learning objective, relative trajectory balance, for training a diffusion model that samples from this posterior, a problem that existing methods solve only approximately or in restricted cases. Relative trajectory balance arises from the generative flow network perspective on diffusion models, which allows the use of deep reinforcement learning techniques to improve mode coverage. Experiments illustrate the broad potential of unbiased inference of arbitrary posteriors under diffusion priors: in vision (classifier guidance), language (infilling under a discrete diffusion LLM), and multimodal data (text-to-image generation). Beyond generative modeling, we apply relative trajectory balance to the problem of continuous control with a score-based behavior prior, achieving state-of-the-art results on benchmarks in offline reinforcement learning. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.20971v2-abstract-full').style.display = 'none'; document.getElementById('2405.20971v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 13 January, 2025; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 31 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">NeurIPS 2024; code: https://github.com/GFNOrg/diffusion-finetuning</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.18540">arXiv:2405.18540</a> <span> [<a href="https://arxiv.org/pdf/2405.18540">pdf</a>, <a href="https://arxiv.org/format/2405.18540">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Cryptography and Security">cs.CR</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Learning diverse attacks on large language models for robust red-teaming and safety tuning </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Lee%2C+S">Seanie Lee</a>, <a href="/search/cs?searchtype=author&query=Kim%2C+M">Minsu Kim</a>, <a href="/search/cs?searchtype=author&query=Cherif%2C+L">Lynn Cherif</a>, <a href="/search/cs?searchtype=author&query=Dobre%2C+D">David Dobre</a>, <a href="/search/cs?searchtype=author&query=Lee%2C+J">Juho Lee</a>, <a href="/search/cs?searchtype=author&query=Hwang%2C+S+J">Sung Ju Hwang</a>, <a href="/search/cs?searchtype=author&query=Kawaguchi%2C+K">Kenji Kawaguchi</a>, <a href="/search/cs?searchtype=author&query=Gidel%2C+G">Gauthier Gidel</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Jain%2C+M">Moksh Jain</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.18540v1-abstract-short" style="display: inline;"> Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that e… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.18540v1-abstract-full').style.display = 'inline'; document.getElementById('2405.18540v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.18540v1-abstract-full" style="display: none;"> Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.18540v1-abstract-full').style.display = 'none'; document.getElementById('2405.18540v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 28 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.13956">arXiv:2405.13956</a> <span> [<a href="https://arxiv.org/pdf/2405.13956">pdf</a>, <a href="https://arxiv.org/format/2405.13956">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Attention as an RNN </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Feng%2C+L">Leo Feng</a>, <a href="/search/cs?searchtype=author&query=Tung%2C+F">Frederick Tung</a>, <a href="/search/cs?searchtype=author&query=Hajimirsadeghi%2C+H">Hossein Hajimirsadeghi</a>, <a href="/search/cs?searchtype=author&query=Ahmed%2C+M+O">Mohamed Osama Ahmed</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Mori%2C+G">Greg Mori</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.13956v2-abstract-short" style="display: inline;"> The advent of Transformers marked a significant breakthrough in sequence modelling, providing a highly performant architecture capable of leveraging GPU parallelism. However, Transformers are computationally expensive at inference time, limiting their applications, particularly in low-resource settings (e.g., mobile and embedded devices). Addressing this, we (1) begin by showing that attention can… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.13956v2-abstract-full').style.display = 'inline'; document.getElementById('2405.13956v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.13956v2-abstract-full" style="display: none;"> The advent of Transformers marked a significant breakthrough in sequence modelling, providing a highly performant architecture capable of leveraging GPU parallelism. However, Transformers are computationally expensive at inference time, limiting their applications, particularly in low-resource settings (e.g., mobile and embedded devices). Addressing this, we (1) begin by showing that attention can be viewed as a special Recurrent Neural Network (RNN) with the ability to compute its \textit{many-to-one} RNN output efficiently. We then (2) show that popular attention-based models such as Transformers can be viewed as RNN variants. However, unlike traditional RNNs (e.g., LSTMs), these models cannot be updated efficiently with new tokens, an important property in sequence modelling. Tackling this, we (3) introduce a new efficient method of computing attention's \textit{many-to-many} RNN output based on the parallel prefix scan algorithm. Building on the new attention formulation, we (4) introduce \textbf{Aaren}, an attention-based module that can not only (i) be trained in parallel (like Transformers) but also (ii) be updated efficiently with new tokens, requiring only constant memory for inferences (like traditional RNNs). Empirically, we show Aarens achieve comparable performance to Transformers on $38$ datasets spread across four popular sequential problem settings: reinforcement learning, event forecasting, time series classification, and time series forecasting tasks while being more time and memory-efficient. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.13956v2-abstract-full').style.display = 'none'; document.getElementById('2405.13956v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 28 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 22 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.13012">arXiv:2405.13012</a> <span> [<a href="https://arxiv.org/pdf/2405.13012">pdf</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Divergent Creativity in Humans and Large Language Models </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Bellemare-Pepin%2C+A">Antoine Bellemare-Pepin</a>, <a href="/search/cs?searchtype=author&query=Lespinasse%2C+F">Fran莽ois Lespinasse</a>, <a href="/search/cs?searchtype=author&query=Th%C3%B6lke%2C+P">Philipp Th枚lke</a>, <a href="/search/cs?searchtype=author&query=Harel%2C+Y">Yann Harel</a>, <a href="/search/cs?searchtype=author&query=Mathewson%2C+K">Kory Mathewson</a>, <a href="/search/cs?searchtype=author&query=Olson%2C+J+A">Jay A. Olson</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Jerbi%2C+K">Karim Jerbi</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.13012v1-abstract-short" style="display: inline;"> The recent surge in the capabilities of Large Language Models (LLMs) has led to claims that they are approaching a level of creativity akin to human capabilities. This idea has sparked a blend of excitement and apprehension. However, a critical piece that has been missing in this discourse is a systematic evaluation of LLM creativity, particularly in comparison to human divergent thinking. To brid… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.13012v1-abstract-full').style.display = 'inline'; document.getElementById('2405.13012v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.13012v1-abstract-full" style="display: none;"> The recent surge in the capabilities of Large Language Models (LLMs) has led to claims that they are approaching a level of creativity akin to human capabilities. This idea has sparked a blend of excitement and apprehension. However, a critical piece that has been missing in this discourse is a systematic evaluation of LLM creativity, particularly in comparison to human divergent thinking. To bridge this gap, we leverage recent advances in creativity science to build a framework for in-depth analysis of divergent creativity in both state-of-the-art LLMs and a substantial dataset of 100,000 humans. We found evidence suggesting that LLMs can indeed surpass human capabilities in specific creative tasks such as divergent association and creative writing. Our quantitative benchmarking framework opens up new paths for the development of more creative LLMs, but it also encourages more granular inquiries into the distinctive elements that constitute human inventive thought processes, compared to those that can be artificially generated. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.13012v1-abstract-full').style.display = 'none'; document.getElementById('2405.13012v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 13 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">First two and last listed authors are corresponding authors. The first two listed authors contributed equally to this work</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.12205">arXiv:2405.12205</a> <span> [<a href="https://arxiv.org/pdf/2405.12205">pdf</a>, <a href="https://arxiv.org/format/2405.12205">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Metacognitive Capabilities of LLMs: An Exploration in Mathematical Problem Solving </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Didolkar%2C+A">Aniket Didolkar</a>, <a href="/search/cs?searchtype=author&query=Goyal%2C+A">Anirudh Goyal</a>, <a href="/search/cs?searchtype=author&query=Ke%2C+N+R">Nan Rosemary Ke</a>, <a href="/search/cs?searchtype=author&query=Guo%2C+S">Siyuan Guo</a>, <a href="/search/cs?searchtype=author&query=Valko%2C+M">Michal Valko</a>, <a href="/search/cs?searchtype=author&query=Lillicrap%2C+T">Timothy Lillicrap</a>, <a href="/search/cs?searchtype=author&query=Rezende%2C+D">Danilo Rezende</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Mozer%2C+M">Michael Mozer</a>, <a href="/search/cs?searchtype=author&query=Arora%2C+S">Sanjeev Arora</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.12205v1-abstract-short" style="display: inline;"> Metacognitive knowledge refers to humans' intuitive knowledge of their own thinking and reasoning processes. Today's best LLMs clearly possess some reasoning processes. The paper gives evidence that they also have metacognitive knowledge, including ability to name skills and procedures to apply given a task. We explore this primarily in context of math reasoning, developing a prompt-guided interac… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.12205v1-abstract-full').style.display = 'inline'; document.getElementById('2405.12205v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.12205v1-abstract-full" style="display: none;"> Metacognitive knowledge refers to humans' intuitive knowledge of their own thinking and reasoning processes. Today's best LLMs clearly possess some reasoning processes. The paper gives evidence that they also have metacognitive knowledge, including ability to name skills and procedures to apply given a task. We explore this primarily in context of math reasoning, developing a prompt-guided interaction procedure to get a powerful LLM to assign sensible skill labels to math questions, followed by having it perform semantic clustering to obtain coarser families of skill labels. These coarse skill labels look interpretable to humans. To validate that these skill labels are meaningful and relevant to the LLM's reasoning processes we perform the following experiments. (a) We ask GPT-4 to assign skill labels to training questions in math datasets GSM8K and MATH. (b) When using an LLM to solve the test questions, we present it with the full list of skill labels and ask it to identify the skill needed. Then it is presented with randomly selected exemplar solved questions associated with that skill label. This improves accuracy on GSM8k and MATH for several strong LLMs, including code-assisted models. The methodology presented is domain-agnostic, even though this article applies it to math problems. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.12205v1-abstract-full').style.display = 'none'; document.getElementById('2405.12205v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 20 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Preprint. Under review</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.06624">arXiv:2405.06624</a> <span> [<a href="https://arxiv.org/pdf/2405.06624">pdf</a>, <a href="https://arxiv.org/format/2405.06624">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Dalrymple%2C+D+%22">David "davidad" Dalrymple</a>, <a href="/search/cs?searchtype=author&query=Skalse%2C+J">Joar Skalse</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Russell%2C+S">Stuart Russell</a>, <a href="/search/cs?searchtype=author&query=Tegmark%2C+M">Max Tegmark</a>, <a href="/search/cs?searchtype=author&query=Seshia%2C+S">Sanjit Seshia</a>, <a href="/search/cs?searchtype=author&query=Omohundro%2C+S">Steve Omohundro</a>, <a href="/search/cs?searchtype=author&query=Szegedy%2C+C">Christian Szegedy</a>, <a href="/search/cs?searchtype=author&query=Goldhaber%2C+B">Ben Goldhaber</a>, <a href="/search/cs?searchtype=author&query=Ammann%2C+N">Nora Ammann</a>, <a href="/search/cs?searchtype=author&query=Abate%2C+A">Alessandro Abate</a>, <a href="/search/cs?searchtype=author&query=Halpern%2C+J">Joe Halpern</a>, <a href="/search/cs?searchtype=author&query=Barrett%2C+C">Clark Barrett</a>, <a href="/search/cs?searchtype=author&query=Zhao%2C+D">Ding Zhao</a>, <a href="/search/cs?searchtype=author&query=Zhi-Xuan%2C+T">Tan Zhi-Xuan</a>, <a href="/search/cs?searchtype=author&query=Wing%2C+J">Jeannette Wing</a>, <a href="/search/cs?searchtype=author&query=Tenenbaum%2C+J">Joshua Tenenbaum</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.06624v3-abstract-short" style="display: inline;"> Ensuring that AI systems reliably and robustly avoid harmful or dangerous behaviours is a crucial challenge, especially for AI systems with a high degree of autonomy and general intelligence, or systems used in safety-critical contexts. In this paper, we will introduce and define a family of approaches to AI safety, which we will refer to as guaranteed safe (GS) AI. The core feature of these appro… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.06624v3-abstract-full').style.display = 'inline'; document.getElementById('2405.06624v3-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.06624v3-abstract-full" style="display: none;"> Ensuring that AI systems reliably and robustly avoid harmful or dangerous behaviours is a crucial challenge, especially for AI systems with a high degree of autonomy and general intelligence, or systems used in safety-critical contexts. In this paper, we will introduce and define a family of approaches to AI safety, which we will refer to as guaranteed safe (GS) AI. The core feature of these approaches is that they aim to produce AI systems which are equipped with high-assurance quantitative safety guarantees. This is achieved by the interplay of three core components: a world model (which provides a mathematical description of how the AI system affects the outside world), a safety specification (which is a mathematical description of what effects are acceptable), and a verifier (which provides an auditable proof certificate that the AI satisfies the safety specification relative to the world model). We outline a number of approaches for creating each of these three core components, describe the main technical challenges, and suggest a number of potential solutions to them. We also argue for the necessity of this approach to AI safety, and for the inadequacy of the main alternative approaches. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.06624v3-abstract-full').style.display = 'none'; document.getElementById('2405.06624v3-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 8 July, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 10 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2405.01616">arXiv:2405.01616</a> <span> [<a href="https://arxiv.org/pdf/2405.01616">pdf</a>, <a href="https://arxiv.org/format/2405.01616">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Biomolecules">q-bio.BM</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Generative Active Learning for the Search of Small-molecule Protein Binders </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Korablyov%2C+M">Maksym Korablyov</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+C">Cheng-Hao Liu</a>, <a href="/search/cs?searchtype=author&query=Jain%2C+M">Moksh Jain</a>, <a href="/search/cs?searchtype=author&query=van+der+Sloot%2C+A+M">Almer M. van der Sloot</a>, <a href="/search/cs?searchtype=author&query=Jolicoeur%2C+E">Eric Jolicoeur</a>, <a href="/search/cs?searchtype=author&query=Ruediger%2C+E">Edward Ruediger</a>, <a href="/search/cs?searchtype=author&query=Nica%2C+A+C">Andrei Cristian Nica</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+E">Emmanuel Bengio</a>, <a href="/search/cs?searchtype=author&query=Lapchevskyi%2C+K">Kostiantyn Lapchevskyi</a>, <a href="/search/cs?searchtype=author&query=St-Cyr%2C+D">Daniel St-Cyr</a>, <a href="/search/cs?searchtype=author&query=Schuetz%2C+D+A">Doris Alexandra Schuetz</a>, <a href="/search/cs?searchtype=author&query=Butoi%2C+V+I">Victor Ion Butoi</a>, <a href="/search/cs?searchtype=author&query=Rector-Brooks%2C+J">Jarrid Rector-Brooks</a>, <a href="/search/cs?searchtype=author&query=Blackburn%2C+S">Simon Blackburn</a>, <a href="/search/cs?searchtype=author&query=Feng%2C+L">Leo Feng</a>, <a href="/search/cs?searchtype=author&query=Nekoei%2C+H">Hadi Nekoei</a>, <a href="/search/cs?searchtype=author&query=Gottipati%2C+S">SaiKrishna Gottipati</a>, <a href="/search/cs?searchtype=author&query=Vijayan%2C+P">Priyesh Vijayan</a>, <a href="/search/cs?searchtype=author&query=Gupta%2C+P">Prateek Gupta</a>, <a href="/search/cs?searchtype=author&query=Ramp%C3%A1%C5%A1ek%2C+L">Ladislav Ramp谩拧ek</a>, <a href="/search/cs?searchtype=author&query=Avancha%2C+S">Sasikanth Avancha</a>, <a href="/search/cs?searchtype=author&query=Bacon%2C+P">Pierre-Luc Bacon</a>, <a href="/search/cs?searchtype=author&query=Hamilton%2C+W+L">William L. Hamilton</a>, <a href="/search/cs?searchtype=author&query=Paige%2C+B">Brooks Paige</a>, <a href="/search/cs?searchtype=author&query=Misra%2C+S">Sanchit Misra</a> , et al. (9 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2405.01616v1-abstract-short" style="display: inline;"> Despite substantial progress in machine learning for scientific discovery in recent years, truly de novo design of small molecules which exhibit a property of interest remains a significant challenge. We introduce LambdaZero, a generative active learning approach to search for synthesizable molecules. Powered by deep reinforcement learning, LambdaZero learns to search over the vast space of molecu… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.01616v1-abstract-full').style.display = 'inline'; document.getElementById('2405.01616v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2405.01616v1-abstract-full" style="display: none;"> Despite substantial progress in machine learning for scientific discovery in recent years, truly de novo design of small molecules which exhibit a property of interest remains a significant challenge. We introduce LambdaZero, a generative active learning approach to search for synthesizable molecules. Powered by deep reinforcement learning, LambdaZero learns to search over the vast space of molecules to discover candidates with a desired property. We apply LambdaZero with molecular docking to design novel small molecules that inhibit the enzyme soluble Epoxide Hydrolase 2 (sEH), while enforcing constraints on synthesizability and drug-likeliness. LambdaZero provides an exponential speedup in terms of the number of calls to the expensive molecular docking oracle, and LambdaZero de novo designed molecules reach docking scores that would otherwise require the virtual screening of a hundred billion molecules. Importantly, LambdaZero discovers novel scaffolds of synthesizable, drug-like inhibitors for sEH. In in vitro experimental validation, a series of ligands from a generated quinazoline-based scaffold were synthesized, and the lead inhibitor N-(4,6-di(pyrrolidin-1-yl)quinazolin-2-yl)-N-methylbenzamide (UM0152893) displayed sub-micromolar enzyme inhibition of sEH. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2405.01616v1-abstract-full').style.display = 'none'; document.getElementById('2405.01616v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 2 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> May 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2404.10094">arXiv:2404.10094</a> <span> [<a href="https://arxiv.org/pdf/2404.10094">pdf</a>, <a href="https://arxiv.org/format/2404.10094">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Quantitative Methods">q-bio.QM</span> </div> </div> <p class="title is-5 mathjax"> Towards DNA-Encoded Library Generation with GFlowNets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Koziarski%2C+M">Micha艂 Koziarski</a>, <a href="/search/cs?searchtype=author&query=Abukalam%2C+M">Mohammed Abukalam</a>, <a href="/search/cs?searchtype=author&query=Shah%2C+V">Vedant Shah</a>, <a href="/search/cs?searchtype=author&query=Vaillancourt%2C+L">Louis Vaillancourt</a>, <a href="/search/cs?searchtype=author&query=Schuetz%2C+D+A">Doris Alexandra Schuetz</a>, <a href="/search/cs?searchtype=author&query=Jain%2C+M">Moksh Jain</a>, <a href="/search/cs?searchtype=author&query=van+der+Sloot%2C+A">Almer van der Sloot</a>, <a href="/search/cs?searchtype=author&query=Bourgey%2C+M">Mathieu Bourgey</a>, <a href="/search/cs?searchtype=author&query=Marinier%2C+A">Anne Marinier</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2404.10094v1-abstract-short" style="display: inline;"> DNA-encoded libraries (DELs) are a powerful approach for rapidly screening large numbers of diverse compounds. One of the key challenges in using DELs is library design, which involves choosing the building blocks that will be combinatorially combined to produce the final library. In this paper we consider the task of protein-protein interaction (PPI) biased DEL design. To this end, we evaluate se… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2404.10094v1-abstract-full').style.display = 'inline'; document.getElementById('2404.10094v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2404.10094v1-abstract-full" style="display: none;"> DNA-encoded libraries (DELs) are a powerful approach for rapidly screening large numbers of diverse compounds. One of the key challenges in using DELs is library design, which involves choosing the building blocks that will be combinatorially combined to produce the final library. In this paper we consider the task of protein-protein interaction (PPI) biased DEL design. To this end, we evaluate several machine learning algorithms on the PPI modulation task and use them as a reward for the proposed GFlowNet-based generative approach. We additionally investigate the possibility of using structural information about building blocks to design a hierarchical action space for the GFlowNet. The observed results indicate that GFlowNets are a promising approach for generating diverse combinatorial library candidates. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2404.10094v1-abstract-full').style.display = 'none'; document.getElementById('2404.10094v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 15 April, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> April 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2404.09932">arXiv:2404.09932</a> <span> [<a href="https://arxiv.org/pdf/2404.09932">pdf</a>, <a href="https://arxiv.org/format/2404.09932">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> </div> </div> <p class="title is-5 mathjax"> Foundational Challenges in Assuring Alignment and Safety of Large Language Models </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Anwar%2C+U">Usman Anwar</a>, <a href="/search/cs?searchtype=author&query=Saparov%2C+A">Abulhair Saparov</a>, <a href="/search/cs?searchtype=author&query=Rando%2C+J">Javier Rando</a>, <a href="/search/cs?searchtype=author&query=Paleka%2C+D">Daniel Paleka</a>, <a href="/search/cs?searchtype=author&query=Turpin%2C+M">Miles Turpin</a>, <a href="/search/cs?searchtype=author&query=Hase%2C+P">Peter Hase</a>, <a href="/search/cs?searchtype=author&query=Lubana%2C+E+S">Ekdeep Singh Lubana</a>, <a href="/search/cs?searchtype=author&query=Jenner%2C+E">Erik Jenner</a>, <a href="/search/cs?searchtype=author&query=Casper%2C+S">Stephen Casper</a>, <a href="/search/cs?searchtype=author&query=Sourbut%2C+O">Oliver Sourbut</a>, <a href="/search/cs?searchtype=author&query=Edelman%2C+B+L">Benjamin L. Edelman</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+Z">Zhaowei Zhang</a>, <a href="/search/cs?searchtype=author&query=G%C3%BCnther%2C+M">Mario G眉nther</a>, <a href="/search/cs?searchtype=author&query=Korinek%2C+A">Anton Korinek</a>, <a href="/search/cs?searchtype=author&query=Hernandez-Orallo%2C+J">Jose Hernandez-Orallo</a>, <a href="/search/cs?searchtype=author&query=Hammond%2C+L">Lewis Hammond</a>, <a href="/search/cs?searchtype=author&query=Bigelow%2C+E">Eric Bigelow</a>, <a href="/search/cs?searchtype=author&query=Pan%2C+A">Alexander Pan</a>, <a href="/search/cs?searchtype=author&query=Langosco%2C+L">Lauro Langosco</a>, <a href="/search/cs?searchtype=author&query=Korbak%2C+T">Tomasz Korbak</a>, <a href="/search/cs?searchtype=author&query=Zhang%2C+H">Heidi Zhang</a>, <a href="/search/cs?searchtype=author&query=Zhong%2C+R">Ruiqi Zhong</a>, <a href="/search/cs?searchtype=author&query=h%C3%89igeartaigh%2C+S+%C3%93">Se谩n 脫 h脡igeartaigh</a>, <a href="/search/cs?searchtype=author&query=Recchia%2C+G">Gabriel Recchia</a>, <a href="/search/cs?searchtype=author&query=Corsi%2C+G">Giulio Corsi</a> , et al. (17 additional authors not shown) </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2404.09932v2-abstract-short" style="display: inline;"> This work identifies 18 foundational challenges in assuring the alignment and safety of large language models (LLMs). These challenges are organized into three different categories: scientific understanding of LLMs, development and deployment methods, and sociotechnical challenges. Based on the identified challenges, we pose $200+$ concrete research questions. </span> <span class="abstract-full has-text-grey-dark mathjax" id="2404.09932v2-abstract-full" style="display: none;"> This work identifies 18 foundational challenges in assuring the alignment and safety of large language models (LLMs). These challenges are organized into three different categories: scientific understanding of LLMs, development and deployment methods, and sociotechnical challenges. Based on the identified challenges, we pose $200+$ concrete research questions. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2404.09932v2-abstract-full').style.display = 'none'; document.getElementById('2404.09932v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 5 September, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 15 April, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> April 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2403.14443">arXiv:2403.14443</a> <span> [<a href="https://arxiv.org/pdf/2403.14443">pdf</a>, <a href="https://arxiv.org/format/2403.14443">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computation and Language">cs.CL</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Computer Science and Game Theory">cs.GT</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Multiagent Systems">cs.MA</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Social and Information Networks">cs.SI</span> </div> </div> <p class="title is-5 mathjax"> Language Models Can Reduce Asymmetry in Information Markets </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Rahaman%2C+N">Nasim Rahaman</a>, <a href="/search/cs?searchtype=author&query=Weiss%2C+M">Martin Weiss</a>, <a href="/search/cs?searchtype=author&query=W%C3%BCthrich%2C+M">Manuel W眉thrich</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Li%2C+L+E">Li Erran Li</a>, <a href="/search/cs?searchtype=author&query=Pal%2C+C">Chris Pal</a>, <a href="/search/cs?searchtype=author&query=Sch%C3%B6lkopf%2C+B">Bernhard Sch枚lkopf</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2403.14443v1-abstract-short" style="display: inline;"> This work addresses the buyer's inspection paradox for information markets. The paradox is that buyers need to access information to determine its value, while sellers need to limit access to prevent theft. To study this, we introduce an open-source simulated digital marketplace where intelligent agents, powered by language models, buy and sell information on behalf of external participants. The c… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.14443v1-abstract-full').style.display = 'inline'; document.getElementById('2403.14443v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2403.14443v1-abstract-full" style="display: none;"> This work addresses the buyer's inspection paradox for information markets. The paradox is that buyers need to access information to determine its value, while sellers need to limit access to prevent theft. To study this, we introduce an open-source simulated digital marketplace where intelligent agents, powered by language models, buy and sell information on behalf of external participants. The central mechanism enabling this marketplace is the agents' dual capabilities: they not only have the capacity to assess the quality of privileged information but also come equipped with the ability to forget. This ability to induce amnesia allows vendors to grant temporary access to proprietary information, significantly reducing the risk of unauthorized retention while enabling agents to accurately gauge the information's relevance to specific queries or tasks. To perform well, agents must make rational decisions, strategically explore the marketplace through generated sub-queries, and synthesize answers from purchased information. Concretely, our experiments (a) uncover biases in language models leading to irrational behavior and evaluate techniques to mitigate these biases, (b) investigate how price affects demand in the context of informational goods, and (c) show that inspection and higher budgets both lead to higher quality outcomes. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.14443v1-abstract-full').style.display = 'none'; document.getElementById('2403.14443v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 21 March, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> March 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2403.07041">arXiv:2403.07041</a> <span> [<a href="https://arxiv.org/pdf/2403.07041">pdf</a>, <a href="https://arxiv.org/format/2403.07041">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Neural and Evolutionary Computing">cs.NE</span> </div> </div> <p class="title is-5 mathjax"> Ant Colony Sampling with GFlowNets for Combinatorial Optimization </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Kim%2C+M">Minsu Kim</a>, <a href="/search/cs?searchtype=author&query=Choi%2C+S">Sanghyeok Choi</a>, <a href="/search/cs?searchtype=author&query=Kim%2C+H">Hyeonah Kim</a>, <a href="/search/cs?searchtype=author&query=Son%2C+J">Jiwoo Son</a>, <a href="/search/cs?searchtype=author&query=Park%2C+J">Jinkyoo Park</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2403.07041v3-abstract-short" style="display: inline;"> We present the Generative Flow Ant Colony Sampler (GFACS), a novel meta-heuristic method that hierarchically combines amortized inference and parallel stochastic search. Our method first leverages Generative Flow Networks (GFlowNets) to amortize a multi-modal prior distribution over combinatorial solution space that encompasses both high-reward and diversified solutions. This prior is iteratively… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.07041v3-abstract-full').style.display = 'inline'; document.getElementById('2403.07041v3-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2403.07041v3-abstract-full" style="display: none;"> We present the Generative Flow Ant Colony Sampler (GFACS), a novel meta-heuristic method that hierarchically combines amortized inference and parallel stochastic search. Our method first leverages Generative Flow Networks (GFlowNets) to amortize a multi-modal prior distribution over combinatorial solution space that encompasses both high-reward and diversified solutions. This prior is iteratively updated via parallel stochastic search in the spirit of Ant Colony Optimization (ACO), leading to the posterior distribution that generates near-optimal solutions. Extensive experiments across seven combinatorial optimization problems demonstrate GFACS's promising performances. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.07041v3-abstract-full').style.display = 'none'; document.getElementById('2403.07041v3-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 11 October, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 11 March, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> March 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">20 pages, 6 figures</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2403.04571">arXiv:2403.04571</a> <span> [<a href="https://arxiv.org/pdf/2403.04571">pdf</a>, <a href="https://arxiv.org/ps/2403.04571">ps</a>, <a href="https://arxiv.org/format/2403.04571">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Artificial Intelligence">cs.AI</span> </div> </div> <p class="title is-5 mathjax"> Machine learning and information theory concepts towards an AI Mathematician </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2403.04571v1-abstract-short" style="display: inline;"> The current state-of-the-art in artificial intelligence is impressive, especially in terms of mastery of language, but not so much in terms of mathematical reasoning. What could be missing? Can we learn something useful about that gap from how the brains of mathematicians go about their craft? This essay builds on the idea that current deep learning mostly succeeds at system 1 abilities -- which c… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.04571v1-abstract-full').style.display = 'inline'; document.getElementById('2403.04571v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2403.04571v1-abstract-full" style="display: none;"> The current state-of-the-art in artificial intelligence is impressive, especially in terms of mastery of language, but not so much in terms of mathematical reasoning. What could be missing? Can we learn something useful about that gap from how the brains of mathematicians go about their craft? This essay builds on the idea that current deep learning mostly succeeds at system 1 abilities -- which correspond to our intuition and habitual behaviors -- but still lacks something important regarding system 2 abilities -- which include reasoning and robust uncertainty estimation. It takes an information-theoretical posture to ask questions about what constitutes an interesting mathematical statement, which could guide future work in crafting an AI mathematician. The focus is not on proving a given theorem but on discovering new and interesting conjectures. The central hypothesis is that a desirable body of theorems better summarizes the set of all provable statements, for example by having a small description length while at the same time being close (in terms of number of derivation steps) to many provable statements. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2403.04571v1-abstract-full').style.display = 'none'; document.getElementById('2403.04571v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 7 March, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> March 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">To appear in the Bulletin of the AMS, 2024</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2402.10309">arXiv:2402.10309</a> <span> [<a href="https://arxiv.org/pdf/2402.10309">pdf</a>, <a href="https://arxiv.org/format/2402.10309">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> </div> </div> <p class="title is-5 mathjax"> Discrete Probabilistic Inference as Control in Multi-path Environments </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Deleu%2C+T">Tristan Deleu</a>, <a href="/search/cs?searchtype=author&query=Nouri%2C+P">Padideh Nouri</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Precup%2C+D">Doina Precup</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2402.10309v2-abstract-short" style="display: inline;"> We consider the problem of sampling from a discrete and structured distribution as a sequential decision problem, where the objective is to find a stochastic policy such that objects are sampled at the end of this sequential process proportionally to some predefined reward. While we could use maximum entropy Reinforcement Learning (MaxEnt RL) to solve this problem for some distributions, it has be… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.10309v2-abstract-full').style.display = 'inline'; document.getElementById('2402.10309v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2402.10309v2-abstract-full" style="display: none;"> We consider the problem of sampling from a discrete and structured distribution as a sequential decision problem, where the objective is to find a stochastic policy such that objects are sampled at the end of this sequential process proportionally to some predefined reward. While we could use maximum entropy Reinforcement Learning (MaxEnt RL) to solve this problem for some distributions, it has been shown that in general, the distribution over states induced by the optimal policy may be biased in cases where there are multiple ways to generate the same object. To address this issue, Generative Flow Networks (GFlowNets) learn a stochastic policy that samples objects proportionally to their reward by approximately enforcing a conservation of flows across the whole Markov Decision Process (MDP). In this paper, we extend recent methods correcting the reward in order to guarantee that the marginal distribution induced by the optimal MaxEnt RL policy is proportional to the original reward, regardless of the structure of the underlying MDP. We also prove that some flow-matching objectives found in the GFlowNet literature are in fact equivalent to well-established MaxEnt RL algorithms with a corrected reward. Finally, we study empirically the performance of multiple MaxEnt RL and GFlowNet algorithms on multiple problems involving sampling from discrete distributions. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.10309v2-abstract-full').style.display = 'none'; document.getElementById('2402.10309v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 27 May, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 15 February, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2024. </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2402.08797">arXiv:2402.08797</a> <span> [<a href="https://arxiv.org/pdf/2402.08797">pdf</a>, <a href="https://arxiv.org/format/2402.08797">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Computers and Society">cs.CY</span> </div> </div> <p class="title is-5 mathjax"> Computing Power and the Governance of Artificial Intelligence </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Sastry%2C+G">Girish Sastry</a>, <a href="/search/cs?searchtype=author&query=Heim%2C+L">Lennart Heim</a>, <a href="/search/cs?searchtype=author&query=Belfield%2C+H">Haydn Belfield</a>, <a href="/search/cs?searchtype=author&query=Anderljung%2C+M">Markus Anderljung</a>, <a href="/search/cs?searchtype=author&query=Brundage%2C+M">Miles Brundage</a>, <a href="/search/cs?searchtype=author&query=Hazell%2C+J">Julian Hazell</a>, <a href="/search/cs?searchtype=author&query=O%27Keefe%2C+C">Cullen O'Keefe</a>, <a href="/search/cs?searchtype=author&query=Hadfield%2C+G+K">Gillian K. Hadfield</a>, <a href="/search/cs?searchtype=author&query=Ngo%2C+R">Richard Ngo</a>, <a href="/search/cs?searchtype=author&query=Pilz%2C+K">Konstantin Pilz</a>, <a href="/search/cs?searchtype=author&query=Gor%2C+G">George Gor</a>, <a href="/search/cs?searchtype=author&query=Bluemke%2C+E">Emma Bluemke</a>, <a href="/search/cs?searchtype=author&query=Shoker%2C+S">Sarah Shoker</a>, <a href="/search/cs?searchtype=author&query=Egan%2C+J">Janet Egan</a>, <a href="/search/cs?searchtype=author&query=Trager%2C+R+F">Robert F. Trager</a>, <a href="/search/cs?searchtype=author&query=Avin%2C+S">Shahar Avin</a>, <a href="/search/cs?searchtype=author&query=Weller%2C+A">Adrian Weller</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Coyle%2C+D">Diane Coyle</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2402.08797v1-abstract-short" style="display: inline;"> Computing power, or "compute," is crucial for the development and deployment of artificial intelligence (AI) capabilities. As a result, governments and companies have started to leverage compute as a means to govern AI. For example, governments are investing in domestic compute capacity, controlling the flow of compute to competing countries, and subsidizing compute access to certain sectors. Howe… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.08797v1-abstract-full').style.display = 'inline'; document.getElementById('2402.08797v1-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2402.08797v1-abstract-full" style="display: none;"> Computing power, or "compute," is crucial for the development and deployment of artificial intelligence (AI) capabilities. As a result, governments and companies have started to leverage compute as a means to govern AI. For example, governments are investing in domestic compute capacity, controlling the flow of compute to competing countries, and subsidizing compute access to certain sectors. However, these efforts only scratch the surface of how compute can be used to govern AI development and deployment. Relative to other key inputs to AI (data and algorithms), AI-relevant compute is a particularly effective point of intervention: it is detectable, excludable, and quantifiable, and is produced via an extremely concentrated supply chain. These characteristics, alongside the singular importance of compute for cutting-edge AI models, suggest that governing compute can contribute to achieving common policy objectives, such as ensuring the safety and beneficial use of AI. More precisely, policymakers could use compute to facilitate regulatory visibility of AI, allocate resources to promote beneficial outcomes, and enforce restrictions against irresponsible or malicious AI development and usage. However, while compute-based policies and technologies have the potential to assist in these areas, there is significant variation in their readiness for implementation. Some ideas are currently being piloted, while others are hindered by the need for fundamental research. Furthermore, naive or poorly scoped approaches to compute governance carry significant risks in areas like privacy, economic impacts, and centralization of power. We end by suggesting guardrails to minimize these risks from compute governance. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.08797v1-abstract-full').style.display = 'none'; document.getElementById('2402.08797v1-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 13 February, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Figures can be accessed at: https://github.com/lheim/CPGAI-Figures</span> </p> </li> <li class="arxiv-result"> <div class="is-marginless"> <p class="list-title is-inline-block"><a href="https://arxiv.org/abs/2402.06121">arXiv:2402.06121</a> <span> [<a href="https://arxiv.org/pdf/2402.06121">pdf</a>, <a href="https://arxiv.org/format/2402.06121">other</a>] </span> </p> <div class="tags is-inline-block"> <span class="tag is-small is-link tooltip is-tooltip-top" data-tooltip="Machine Learning">cs.LG</span> <span class="tag is-small is-grey tooltip is-tooltip-top" data-tooltip="Machine Learning">stat.ML</span> </div> </div> <p class="title is-5 mathjax"> Iterated Denoising Energy Matching for Sampling from Boltzmann Densities </p> <p class="authors"> <span class="search-hit">Authors:</span> <a href="/search/cs?searchtype=author&query=Akhound-Sadegh%2C+T">Tara Akhound-Sadegh</a>, <a href="/search/cs?searchtype=author&query=Rector-Brooks%2C+J">Jarrid Rector-Brooks</a>, <a href="/search/cs?searchtype=author&query=Bose%2C+A+J">Avishek Joey Bose</a>, <a href="/search/cs?searchtype=author&query=Mittal%2C+S">Sarthak Mittal</a>, <a href="/search/cs?searchtype=author&query=Lemos%2C+P">Pablo Lemos</a>, <a href="/search/cs?searchtype=author&query=Liu%2C+C">Cheng-Hao Liu</a>, <a href="/search/cs?searchtype=author&query=Sendera%2C+M">Marcin Sendera</a>, <a href="/search/cs?searchtype=author&query=Ravanbakhsh%2C+S">Siamak Ravanbakhsh</a>, <a href="/search/cs?searchtype=author&query=Gidel%2C+G">Gauthier Gidel</a>, <a href="/search/cs?searchtype=author&query=Bengio%2C+Y">Yoshua Bengio</a>, <a href="/search/cs?searchtype=author&query=Malkin%2C+N">Nikolay Malkin</a>, <a href="/search/cs?searchtype=author&query=Tong%2C+A">Alexander Tong</a> </p> <p class="abstract mathjax"> <span class="has-text-black-bis has-text-weight-semibold">Abstract</span>: <span class="abstract-short has-text-grey-dark mathjax" id="2402.06121v2-abstract-short" style="display: inline;"> Efficiently generating statistically independent samples from an unnormalized probability distribution, such as equilibrium samples of many-body systems, is a foundational problem in science. In this paper, we propose Iterated Denoising Energy Matching (iDEM), an iterative algorithm that uses a novel stochastic score matching objective leveraging solely the energy function and its gradient -- and… <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.06121v2-abstract-full').style.display = 'inline'; document.getElementById('2402.06121v2-abstract-short').style.display = 'none';">▽ More</a> </span> <span class="abstract-full has-text-grey-dark mathjax" id="2402.06121v2-abstract-full" style="display: none;"> Efficiently generating statistically independent samples from an unnormalized probability distribution, such as equilibrium samples of many-body systems, is a foundational problem in science. In this paper, we propose Iterated Denoising Energy Matching (iDEM), an iterative algorithm that uses a novel stochastic score matching objective leveraging solely the energy function and its gradient -- and no data samples -- to train a diffusion-based sampler. Specifically, iDEM alternates between (I) sampling regions of high model density from a diffusion-based sampler and (II) using these samples in our stochastic matching objective to further improve the sampler. iDEM is scalable to high dimensions as the inner matching objective, is simulation-free, and requires no MCMC samples. Moreover, by leveraging the fast mode mixing behavior of diffusion, iDEM smooths out the energy landscape enabling efficient exploration and learning of an amortized sampler. We evaluate iDEM on a suite of tasks ranging from standard synthetic energy functions to invariant $n$-body particle systems. We show that the proposed approach achieves state-of-the-art performance on all metrics and trains $2-5\times$ faster, which allows it to be the first method to train using energy on the challenging $55$-particle Lennard-Jones system. <a class="is-size-7" style="white-space: nowrap;" onclick="document.getElementById('2402.06121v2-abstract-full').style.display = 'none'; document.getElementById('2402.06121v2-abstract-short').style.display = 'inline';">△ Less</a> </span> </p> <p class="is-size-7"><span class="has-text-black-bis has-text-weight-semibold">Submitted</span> 26 June, 2024; <span class="has-text-black-bis has-text-weight-semibold">v1</span> submitted 8 February, 2024; <span class="has-text-black-bis has-text-weight-semibold">originally announced</span> February 2024. </p> <p class="comments is-size-7"> <span class="has-text-black-bis has-text-weight-semibold">Comments:</span> <span class="has-text-grey-dark mathjax">Published at ICML 2024. Code for iDEM is available at https://github.com/jarridrb/dem</span> </p> </li> </ol> <nav class="pagination is-small is-centered breathe-horizontal" role="navigation" aria-label="pagination"> <a href="" class="pagination-previous is-invisible">Previous </a> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=50" class="pagination-next" >Next </a> <ul class="pagination-list"> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=0" class="pagination-link is-current" aria-label="Goto page 1">1 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=50" class="pagination-link " aria-label="Page 2" aria-current="page">2 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=100" class="pagination-link " aria-label="Page 3" aria-current="page">3 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=150" class="pagination-link " aria-label="Page 4" aria-current="page">4 </a> </li> <li> <a href="/search/?searchtype=author&query=Bengio%2C+Y&start=200" class="pagination-link " aria-label="Page 5" aria-current="page">5 </a> </li> <li><span class="pagination-ellipsis">…</span></li> </ul> </nav> <div class="is-hidden-tablet">  <span class="help" style="display: inline-block;"><a href="https://github.com/arXiv/arxiv-search/releases">Search v0.5.6 released 2020-02-24</a>  </span> </div> </div> </main> <footer> <div class="columns is-desktop" role="navigation" aria-label="Secondary">  <div class="column"> <div class="columns"> <div class="column"> <ul class="nav-spaced"> <li><a href="https://info.arxiv.org/about">About</a></li> <li><a href="https://info.arxiv.org/help">Help</a></li> </ul> </div> <div class="column"> <ul class="nav-spaced"> <li> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512" class="icon filter-black" role="presentation"><title>contact arXiv</title><desc>Click here to contact arXiv</desc><path d="M502.3 190.8c3.9-3.1 9.7-.2 9.7 4.7V400c0 26.5-21.5 48-48 48H48c-26.5 0-48-21.5-48-48V195.6c0-5 5.7-7.8 9.7-4.7 22.4 17.4 52.1 39.5 154.1 113.6 21.1 15.4 56.7 47.8 92.2 47.6 35.7.3 72-32.8 92.3-47.6 102-74.1 131.6-96.3 154-113.7zM256 320c23.2.4 56.6-29.2 73.4-41.4 132.7-96.3 142.8-104.7 173.4-128.7 5.8-4.5 9.2-11.5 9.2-18.9v-19c0-26.5-21.5-48-48-48H48C21.5 64 0 85.5 0 112v19c0 7.4 3.4 14.3 9.2 18.9 30.6 23.9 40.7 32.4 173.4 128.7 16.8 12.2 50.2 41.8 73.4 41.4z"/></svg> <a href="https://info.arxiv.org/help/contact.html"> Contact</a> </li> <li> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512" class="icon filter-black" role="presentation"><title>subscribe to arXiv mailings</title><desc>Click here to subscribe</desc><path d="M476 3.2L12.5 270.6c-18.1 10.4-15.8 35.6 2.2 43.2L121 358.4l287.3-253.2c5.5-4.9 13.3 2.6 8.6 8.3L176 407v80.5c0 23.6 28.5 32.9 42.5 15.8L282 426l124.6 52.2c14.2 6 30.4-2.9 33-18.2l72-432C515 7.8 493.3-6.8 476 3.2z"/></svg> <a href="https://info.arxiv.org/help/subscribe"> Subscribe</a> </li> </ul> </div> </div> </div>   <div class="column"> <div class="columns"> <div class="column"> <ul class="nav-spaced"> <li><a href="https://info.arxiv.org/help/license/index.html">Copyright</a></li> <li><a href="https://info.arxiv.org/help/policies/privacy_policy.html">Privacy Policy</a></li> </ul> </div> <div class="column sorry-app-links"> <ul class="nav-spaced"> <li><a href="https://info.arxiv.org/help/web_accessibility.html">Web Accessibility Assistance</a></li> <li> <p class="help"> <a class="a11y-main-link" href="https://status.arxiv.org" target="_blank">arXiv Operational Status <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 256 512" class="icon filter-dark_grey" role="presentation"><path d="M224.3 273l-136 136c-9.4 9.4-24.6 9.4-33.9 0l-22.6-22.6c-9.4-9.4-9.4-24.6 0-33.9l96.4-96.4-96.4-96.4c-9.4-9.4-9.4-24.6 0-33.9L54.3 103c9.4-9.4 24.6-9.4 33.9 0l136 136c9.5 9.4 9.5 24.6.1 34z"/></svg></a><br> Get status notifications via <a class="is-link" href="https://subscribe.sorryapp.com/24846f03/email/new" target="_blank"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512" class="icon filter-black" role="presentation"><path d="M502.3 190.8c3.9-3.1 9.7-.2 9.7 4.7V400c0 26.5-21.5 48-48 48H48c-26.5 0-48-21.5-48-48V195.6c0-5 5.7-7.8 9.7-4.7 22.4 17.4 52.1 39.5 154.1 113.6 21.1 15.4 56.7 47.8 92.2 47.6 35.7.3 72-32.8 92.3-47.6 102-74.1 131.6-96.3 154-113.7zM256 320c23.2.4 56.6-29.2 73.4-41.4 132.7-96.3 142.8-104.7 173.4-128.7 5.8-4.5 9.2-11.5 9.2-18.9v-19c0-26.5-21.5-48-48-48H48C21.5 64 0 85.5 0 112v19c0 7.4 3.4 14.3 9.2 18.9 30.6 23.9 40.7 32.4 173.4 128.7 16.8 12.2 50.2 41.8 73.4 41.4z"/></svg>email</a> or <a class="is-link" href="https://subscribe.sorryapp.com/24846f03/slack/new" target="_blank"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 448 512" class="icon filter-black" role="presentation"><path d="M94.12 315.1c0 25.9-21.16 47.06-47.06 47.06S0 341 0 315.1c0-25.9 21.16-47.06 47.06-47.06h47.06v47.06zm23.72 0c0-25.9 21.16-47.06 47.06-47.06s47.06 21.16 47.06 47.06v117.84c0 25.9-21.16 47.06-47.06 47.06s-47.06-21.16-47.06-47.06V315.1zm47.06-188.98c-25.9 0-47.06-21.16-47.06-47.06S139 32 164.9 32s47.06 21.16 47.06 47.06v47.06H164.9zm0 23.72c25.9 0 47.06 21.16 47.06 47.06s-21.16 47.06-47.06 47.06H47.06C21.16 243.96 0 222.8 0 196.9s21.16-47.06 47.06-47.06H164.9zm188.98 47.06c0-25.9 21.16-47.06 47.06-47.06 25.9 0 47.06 21.16 47.06 47.06s-21.16 47.06-47.06 47.06h-47.06V196.9zm-23.72 0c0 25.9-21.16 47.06-47.06 47.06-25.9 0-47.06-21.16-47.06-47.06V79.06c0-25.9 21.16-47.06 47.06-47.06 25.9 0 47.06 21.16 47.06 47.06V196.9zM283.1 385.88c25.9 0 47.06 21.16 47.06 47.06 0 25.9-21.16 47.06-47.06 47.06-25.9 0-47.06-21.16-47.06-47.06v-47.06h47.06zm0-23.72c-25.9 0-47.06-21.16-47.06-47.06 0-25.9 21.16-47.06 47.06-47.06h117.84c25.9 0 47.06 21.16 47.06 47.06 0 25.9-21.16 47.06-47.06 47.06H283.1z"/></svg>slack</a> </p> </li> </ul> </div> </div> </div>  </div> </footer> <script src="https://static.arxiv.org/static/base/1.0.0a5/js/member_acknowledgement.js"></script> </body> </html>

CINXE.COM

Search | arXiv e-print repository