generated from alshedivat/al-folio
-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
1 lines (1 loc) · 20.4 KB
/
index.html
1
<!DOCTYPE html> <html lang="en"> <head> <meta http-equiv="Content-Type" content="text/html; charset=UTF-8"> <meta charset="utf-8"> <meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no"> <meta http-equiv="X-UA-Compatible" content="IE=edge"> <title> Louis Grenioux </title> <meta name="author" content="Louis Grenioux"> <meta name="description" content="Personal website of Louis Grenioux "> <meta name="keywords" content="h2o64, sampling, energy-based models, ebm, normalizing flows, nf, flow matching, fm, mcmc"> <link rel="stylesheet" href="/assets/css/bootstrap.min.css?a4b3f509e79c54a512b890d73235ef04"> <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/css/mdb.min.css" integrity="sha256-jpjYvU3G3N6nrrBwXJoVEYI/0zw8htfFnhT9ljN3JJw=" crossorigin="anonymous"> <link defer rel="stylesheet" href="/assets/css/academicons.min.css?f0b7046b84e425c55f3463ac249818f5"> <link defer rel="stylesheet" type="text/css" href="https://fonts.googleapis.com/css?family=Roboto:300,400,500,700|Roboto+Slab:100,300,400,500,700|Material+Icons&display=swap"> <link defer rel="stylesheet" href="/assets/css/jekyll-pygments-themes-github.css?591dab5a4e56573bf4ef7fd332894c99" media="" id="highlight_theme_light"> <link rel="shortcut icon" href="/assets/img/favicon.ico?20ce5773dab6ac2faaf9d85606a581b3"> <link rel="stylesheet" href="/assets/css/main.css?d41d8cd98f00b204e9800998ecf8427e"> <link rel="canonical" href="https://h2o64.github.io/"> <script src="/assets/js/theme.js?9a0c749ec5240d9cda97bc72359a72c0"></script> <link defer rel="stylesheet" href="/assets/css/jekyll-pygments-themes-native.css?5847e5ed4a4568527aa6cfab446049ca" media="none" id="highlight_theme_dark"> <script>initTheme();</script> </head> <body class="fixed-top-nav sticky-bottom-footer"> <header> <nav id="navbar" class="navbar navbar-light navbar-expand-sm fixed-top" role="navigation"> <div class="container"> <button class="navbar-toggler collapsed ml-auto" type="button" data-toggle="collapse" data-target="#navbarNav" aria-controls="navbarNav" aria-expanded="false" aria-label="Toggle navigation"> <span class="sr-only">Toggle navigation</span> <span class="icon-bar top-bar"></span> <span class="icon-bar middle-bar"></span> <span class="icon-bar bottom-bar"></span> </button> <div class="collapse navbar-collapse text-right" id="navbarNav"> <ul class="navbar-nav ml-auto flex-nowrap"> <li class="nav-item active"> <a class="nav-link" href="/">about me <span class="sr-only">(current)</span> </a> </li> <li class="nav-item "> <a class="nav-link" href="/publications/">publications </a> </li> <li class="nav-item "> <a class="nav-link" href="/cv/">cv </a> </li> <li class="toggle-container"> <button id="light-toggle" title="Change theme"> <i class="ti ti-sun-moon" id="light-toggle-system"></i> <i class="ti ti-moon-filled" id="light-toggle-dark"></i> <i class="ti ti-sun-filled" id="light-toggle-light"></i> </button> </li> </ul> </div> </div> </nav> <progress id="progress" value="0"> <div class="progress-container"> <span class="progress-bar"></span> </div> </progress> </header> <div class="container mt-5" role="main"> <div class="post"> <header class="post-header"> <h1 class="post-title"> <span class="font-weight-bold">Louis</span> Grenioux </h1> <p class="desc"></p> </header> <article> <div class="profile float-right"> <figure> <picture> <source class="responsive-img-srcset" srcset="/assets/img/me-480.webp 480w,/assets/img/me-800.webp 800w,/assets/img/me-1400.webp 1400w," sizes="(min-width: 930px) 270.0px, (min-width: 576px) 30vw, 95vw" type="image/webp"> <img src="/assets/img/me.jpg?8d60564e43bedcd5a99452d666865248" class="img-fluid z-depth-1 rounded" width="100%" height="auto" alt="me.jpg" loading="eager" onerror="this.onerror=null; $('.responsive-img-srcset').remove();"> </source></picture> </figure> </div> <div class="clearfix"> <p><strong>About</strong> I’m a PhD student at <a href="https://cmap.ip-paris.fr/" rel="external nofollow noopener" target="_blank">Centre de Mathématiques Appliquées</a> (CMAP) at <a href="https://www.polytechnique.edu/" rel="external nofollow noopener" target="_blank">École polytechnique</a> with <a href="https://marylou-gabrie.github.io/" rel="external nofollow noopener" target="_blank">Marylou Gabrié</a> and <a href="https://scholar.google.fr/citations?user=_XE1LvQAAAAJ" rel="external nofollow noopener" target="_blank">Éric Moulines</a>. I am a gratuated engineer from <a href="https://www.telecom-sudparis.eu/" rel="external nofollow noopener" target="_blank">Télécom SudParis </a>. My reasearch is generously supported by the <a href="https://www.hi-paris.fr/" rel="external nofollow noopener" target="_blank">Hi! Paris research center</a>.</p> <p><strong>Research interests</strong> Generative Models, Sampling, Energy Based Models, Diffusion Models, Flow Matching, Markov Chain Monte Carlo</p> </div> <h2> <a href="/news/" style="color: inherit">news</a> </h2> <div class="news"> <div class="table-responsive" style="max-height: 60vw"> <table class="table table-sm table-borderless"> <tr> <th scope="row" style="width: 20%">May 04, 2024</th> <td> I will be at the <a href="https://www.unive.it/web/en/2208/home" rel="external nofollow noopener" target="_blank">ISBA 2024 World Meeting</a> (Venice, Italy) in July 2024. I will give a talk entitled “Transporting measures for sampling: parametric and non-parametric approaches inspired by generative modeling” in the invited session “Monte Carlo algorithms using modern hardware” chaired by <a href="https://charlesm93.github.io/" rel="external nofollow noopener" target="_blank">Charles Margossian</a> (Flatiron Institute, New York, USA). </td> </tr> <tr> <th scope="row" style="width: 20%">May 01, 2024</th> <td> My latest paper <a href="https://arxiv.org/abs/2402.10758" rel="external nofollow noopener" target="_blank">Stochastic Localization via Iterative Posterior Sampling</a> co-authored with Maxence Noble, Marylou Gabrié and Alain Durmus has been accepted at ICML 2024. </td> </tr> <tr> <th scope="row" style="width: 20%">Apr 19, 2024</th> <td> I will give a talk at the <a href="https://www.bernoulli-ims-worldcongress2024.org/" rel="external nofollow noopener" target="_blank">11th World Congress in Probability and Statistics</a> (Bochum, Germany) in August about sampling with approximate transport maps within the <a href="https://www.bernoulli-ims-worldcongress2024.org/organized-contributed-paper-sessions" rel="external nofollow noopener" target="_blank">New Trends in Approximate Sampling</a> session chaired by <a href="https://staff.fim.uni-passau.de/~rudolf/" rel="external nofollow noopener" target="_blank">Daniel Rudolf</a> (University of Passau, Germany) and <a href="https://scholar.google.de/citations?user=D3XfgpkAAAAJ" rel="external nofollow noopener" target="_blank">Björn Sprungk</a> (University of Freiberg, Germany). </td> </tr> <tr> <th scope="row" style="width: 20%">Sep 04, 2023</th> <td> I will be at Orsay for the <a href="https://indico.ijclab.in2p3.fr/event/9042/" rel="external nofollow noopener" target="_blank">Probabilistic sampling for physics: finding needles in a field of high-dimensional haystacks</a> workshop organized by Institut Pascal from the 4th September to the 22nd September. Moreover, I will be at Marne-la-Vallée for the <a href="https://sites.google.com/view/aleiac/anr-sineq/summer-school-mol-dyn-on-julia" rel="external nofollow noopener" target="_blank">Sampling high-dimensional probability measures: applications in (non)equilibrium molecular dynamics and statistics</a> at the <a href="https://cermics-lab.enpc.fr/" rel="external nofollow noopener" target="_blank">CERMICS</a> lab where I will be presenting 2 posters. </td> </tr> </table> </div> </div> <h2> <a href="/publications/" style="color: inherit">selected publications</a> </h2> <div class="publications"> <ol class="bibliography"> <li> <div class="row"> <div class="col col-sm-2 abbr"> <abbr class="badge rounded w-100">ICML</abbr> </div> <div id="grenioux2024stochastic" class="col-sm-8"> <div class="title">Stochastic Localization via Iterative Posterior Sampling</div> <div class="author"> <em>Louis Grenioux*</em>, <a href="https://scholar.google.com/citations?user=4eGHx3gAAAAJ" rel="external nofollow noopener" target="_blank">Maxence Noble*</a>, <a href="https://marylou-gabrie.github.io/" rel="external nofollow noopener" target="_blank">Marylou Gabrié</a>, and <span class="more-authors" title="click to view 1 more author" onclick=" var element=$(this); element.attr('title', ''); var more_authors_text=element.text() == '1 more author' ? 'Alain Durmus' : '1 more author'; var cursorPosition=0; var textAdder=setInterval(function(){ element.text(more_authors_text.substring(0, cursorPosition + 1)); if (++cursorPosition == more_authors_text.length){ clearInterval(textAdder); } }, '10'); ">1 more author</span> </div> <div class="periodical"> 2024 </div> <div class="periodical"> </div> <div class="links"> <a class="award btn btn-sm z-depth-0" role="button">Spotlight</a> <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> <a href="https://arxiv.org/pdf/2402.10758" class="btn btn-sm z-depth-0" role="button" rel="external nofollow noopener" target="_blank">PDF</a> <a href="https://github.com/h2o64/slips" class="btn btn-sm z-depth-0" role="button" rel="external nofollow noopener" target="_blank">Code</a> </div> <div class="award hidden d-print-inline"> <p></p> <p>This paper has been selected as a spotlight-designated paper at the conference. Top 3.5% acceptance rate.</p> </div> <div class="abstract hidden"> <p>Building upon score-based learning, new interest in stochastic localization techniques has recently emerged. In these models, one seeks to noise a sample from the data distribution through a stochastic process, called observation process, and progressively learns a denoiser associated to this dynamics. Apart from specific applications, the use of stochastic localization for the problem of sampling from an unnormalized target density has not been explored extensively. This work contributes to fill this gap. We consider a general stochastic localization framework and introduce an explicit class of observation processes, associated with flexible denoising schedules. We provide a complete methodology, Stochastic Localization via Iterative Posterior Sampling (SLIPS), to obtain approximate samples of these dynamics, and as a by-product, samples from the target distribution. Our scheme is based on a Markov chain Monte Carlo estimation of the denoiser and comes with detailed practical guidelines. We illustrate the benefits and applicability of SLIPS on several benchmarks, including Gaussian mixtures in increasing dimensions, Bayesian logistic regression and a high-dimensional field system from statistical-mechanics.</p> </div> </div> </div> </li> <li> <div class="row"> <div class="col col-sm-2 abbr"> <abbr class="badge rounded w-100">ICML</abbr> </div> <div id="grenioux2023balanced" class="col-sm-8"> <div class="title">Balanced Training of Energy-Based Models with Adaptive Flow Sampling</div> <div class="author"> <em>Louis Grenioux</em>, <a href="https://scholar.google.fr/citations?user=_XE1LvQAAAAJ" rel="external nofollow noopener" target="_blank">Éric Moulines</a>, and <a href="https://marylou-gabrie.github.io/" rel="external nofollow noopener" target="_blank">Marylou Gabrié</a> </div> <div class="periodical"> <em>Workshop on Structured Probabilistic Inference & Generative Modeling (SPIGM)</em>, 2023 </div> <div class="periodical"> </div> <div class="links"> <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> <a href="https://openreview.net/pdf?id=AwJ2NqxWlk" class="btn btn-sm z-depth-0" role="button" rel="external nofollow noopener" target="_blank">PDF</a> <a href="/assets/pdf/poster_ebm_spigm2023.pdf" class="btn btn-sm z-depth-0" role="button">Poster</a> </div> <div class="abstract hidden"> <p>Energy-based models (EBMs) are versatile density estimation models that directly parameterize an unnormalized log density. Although very flexible, EBMs lack a specified normalization constant of the model, making the likelihood of the model computationally intractable. Several approximate samplers and variational inference techniques have been proposed to estimate the likelihood gradients for training. These techniques have shown promising results in generating samples, but little attention has been paid to the statistical accuracy of the estimated density, such as determining the relative importance of different classes in a dataset. In this work, we propose a new maximum likelihood training algorithm for EBMs that uses a different type of generative model, normalizing flows (NF), which have recently been proposed to facilitate sampling. Our method fits an NF to an EBM during training so that an NF-assisted sampling scheme provides an accurate gradient for the EBMs at all times, ultimately leading to a fast sampler for generating new data.</p> </div> </div> </div> </li> <li> <div class="row"> <div class="col col-sm-2 abbr"> <abbr class="badge rounded w-100">ICML</abbr> </div> <div id="grenioux2023sampling" class="col-sm-8"> <div class="title">On Sampling with Approximate Transport Maps</div> <div class="author"> <em>Louis Grenioux</em>, <a href="http://alain.perso.math.cnrs.fr/" rel="external nofollow noopener" target="_blank">Alain Durmus</a>, <a href="https://scholar.google.fr/citations?user=_XE1LvQAAAAJ" rel="external nofollow noopener" target="_blank">Éric Moulines</a>, and <span class="more-authors" title="click to view 1 more author" onclick=" var element=$(this); element.attr('title', ''); var more_authors_text=element.text() == '1 more author' ? 'Marylou Gabrié' : '1 more author'; var cursorPosition=0; var textAdder=setInterval(function(){ element.text(more_authors_text.substring(0, cursorPosition + 1)); if (++cursorPosition == more_authors_text.length){ clearInterval(textAdder); } }, '10'); ">1 more author</span> </div> <div class="periodical"> 2023 </div> <div class="periodical"> </div> <div class="links"> <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> <a href="https://proceedings.mlr.press/v202/grenioux23a/grenioux23a.pdf" class="btn btn-sm z-depth-0" role="button" rel="external nofollow noopener" target="_blank">PDF</a> <a href="https://github.com/h2o64/flow_mcmc" class="btn btn-sm z-depth-0" role="button" rel="external nofollow noopener" target="_blank">Code</a> <a href="/assets/pdf/poster_icml2023.pdf" class="btn btn-sm z-depth-0" role="button">Poster</a> </div> <div class="abstract hidden"> <p>Transport maps can ease the sampling of distributions with non-trivial geometries by transforming them into distributions that are easier to handle. The potential of this approach has risen with the development of Normalizing Flows (NF) which are maps parameterized with deep neural networks trained to push a reference distribution towards a target. NF-enhanced samplers recently proposed blend (Markov chain) Monte Carlo methods with either (i) proposal draws from the flow or (ii) a flow-based reparametrization. In both cases, the quality of the learned transport conditions performance. The present work clarifies for the first time the relative strengths and weaknesses of these two approaches. Our study concludes that multimodal targets can be reliably handled with flow-based proposals up to moderately high dimensions. In contrast, methods relying on reparametrization struggle with multimodality but are more robust otherwise in high-dimensional settings and under poor training. To further illustrate the influence of target-proposal adequacy, we also derive a new quantitative bound for the mixing time of the Independent Metropolis-Hastings sampler.</p> </div> </div> </div> </li> </ol> </div> <div class="social"> <div class="contact-icons"> <a href="mailto:%6C%6F%75%69%73.%67%72%65%6E%69%6F%75%78@%70%6F%6C%79%74%65%63%68%6E%69%71%75%65.%65%64%75" title="email"><i class="fa-solid fa-envelope"></i></a> <a href="https://telegram.me/h2o64" title="telegram" rel="external nofollow noopener" target="_blank"><i class="fa-brands fa-telegram"></i></a> <a href="https://orcid.org/0000-0002-1864-3595" title="ORCID" rel="external nofollow noopener" target="_blank"><i class="ai ai-orcid"></i></a> <a href="https://scholar.google.com/citations?user=eLpNKSEAAAAJ" title="Google Scholar" rel="external nofollow noopener" target="_blank"><i class="ai ai-google-scholar"></i></a> <a href="https://github.com/h2o64" title="GitHub" rel="external nofollow noopener" target="_blank"><i class="fa-brands fa-github"></i></a> <a href="https://www.linkedin.com/in/lgrenioux" title="LinkedIn" rel="external nofollow noopener" target="_blank"><i class="fa-brands fa-linkedin"></i></a> <a href="https://twitter.com/theh2o64" title="X" rel="external nofollow noopener" target="_blank"><i class="fa-brands fa-x-twitter"></i></a> </div> <div class="contact-note"></div> </div> </article> </div> </div> <footer class="sticky-bottom mt-5" role="contentinfo"> <div class="container"> © Copyright 2024 Louis Grenioux. Powered by <a href="https://jekyllrb.com/" target="_blank" rel="external nofollow noopener">Jekyll</a> with <a href="https://github.com/alshedivat/al-folio" rel="external nofollow noopener" target="_blank">al-folio</a> theme. Hosted by <a href="https://pages.github.com/" target="_blank" rel="external nofollow noopener">GitHub Pages</a>. Last updated: June 17, 2024. </div> </footer> <script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/jquery.min.js" integrity="sha256-/xUj+3OJU5yExlq6GSYGSHk7tPXikynS7ogEvDej/m4=" crossorigin="anonymous"></script> <script src="/assets/js/bootstrap.bundle.min.js"></script> <script src="https://cdn.jsdelivr.net/npm/[email protected]/js/mdb.min.js" integrity="sha256-NdbiivsvWt7VYCt6hYNT3h/th9vSTL4EDWeGs5SN3DA=" crossorigin="anonymous"></script> <script defer src="https://cdn.jsdelivr.net/npm/[email protected]/dist/masonry.pkgd.min.js" integrity="sha256-Nn1q/fx0H7SNLZMQ5Hw5JLaTRZp0yILA/FRexe19VdI=" crossorigin="anonymous"></script> <script defer src="https://cdn.jsdelivr.net/npm/[email protected]/imagesloaded.pkgd.min.js" integrity="sha256-htrLFfZJ6v5udOG+3kNLINIKh2gvoKqwEhHYfTTMICc=" crossorigin="anonymous"></script> <script defer src="/assets/js/masonry.js" type="text/javascript"></script> <script defer src="https://cdn.jsdelivr.net/npm/[email protected]/dist/medium-zoom.min.js" integrity="sha256-ZgMyDAIYDYGxbcpJcfUnYwNevG/xi9OHKaR/8GK+jWc=" crossorigin="anonymous"></script> <script defer src="/assets/js/zoom.js?85ddb88934d28b74e78031fd54cf8308"></script> <script src="/assets/js/no_defer.js?2781658a0a2b13ed609542042a859126"></script> <script defer src="/assets/js/common.js?06cae41083477f121be8cd9797ad8e2f"></script> <script defer src="/assets/js/copy_code.js?12775fdf7f95e901d7119054556e495f" type="text/javascript"></script> <script defer src="/assets/js/jupyter_new_tab.js?d9f17b6adc2311cbabd747f4538bb15f"></script> <script async src="https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js"></script> <script async src="https://badge.dimensions.ai/badge.js"></script> <script type="text/javascript">window.MathJax={tex:{tags:"ams"}};</script> <script defer type="text/javascript" id="MathJax-script" src="https://cdn.jsdelivr.net/npm/[email protected]/es5/tex-mml-chtml.min.js" integrity="sha256-rjmgmaB99riUNcdlrDtcAiwtLIojSxNyUFdl+Qh+rB4=" crossorigin="anonymous"></script> <script defer src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6" crossorigin="anonymous"></script> <script type="text/javascript">function progressBarSetup(){"max"in document.createElement("progress")?(initializeProgressElement(),$(document).on("scroll",function(){progressBar.attr({value:getCurrentScrollPosition()})}),$(window).on("resize",initializeProgressElement)):(resizeProgressBar(),$(document).on("scroll",resizeProgressBar),$(window).on("resize",resizeProgressBar))}function getCurrentScrollPosition(){return $(window).scrollTop()}function initializeProgressElement(){let e=$("#navbar").outerHeight(!0);$("body").css({"padding-top":e}),$("progress-container").css({"padding-top":e}),progressBar.css({top:e}),progressBar.attr({max:getDistanceToScroll(),value:getCurrentScrollPosition()})}function getDistanceToScroll(){return $(document).height()-$(window).height()}function resizeProgressBar(){progressBar.css({width:getWidthPercentage()+"%"})}function getWidthPercentage(){return getCurrentScrollPosition()/getDistanceToScroll()*100}const progressBar=$("#progress");window.onload=function(){setTimeout(progressBarSetup,50)};</script> <script src="/assets/js/vanilla-back-to-top.min.js?f40d453793ff4f64e238e420181a1d17"></script> <script>addBackToTop();</script> <script src="/assets/js/shortcut-key.js"></script> </body> </html>