1 2<html lang="en-US"> 3<!--<![endif]--> 4<head> 5<meta charset="UTF-8" /> 6<meta name="viewport" content="width=device-width" /> 7<title>Le Seminaire Palaisien</title> 8<!--[if lt IE 9]>
9<script src="http://dpkingma.com/wordpress/wp-content/themes/twentytwelve/js/html5.js" type="text/javascript"></script>
9 10<![endif]-->
11<script src="https://use.fontawesome.com/ae904b9937.js"></script>
11 12 13 14 15<link rel="shortcut icon" href="/static/programme/images/favicon.png"> 16<link rel="stylesheet" type="text/css" href="/static/programme/css/style.css" /> 17<link rel="stylesheet" href="/static/programme/css/academicons.css"/>
18<script type="text/javascript" src="/static/programme/javascript/lib.js"></script>
18 19<!-- highlight current page in navigation -->
20<script> 21$(function(){ 22 $('a').each(function() { 23 if ($(this).prop('href') == window.location.href) { 24 $(this).parent().addClass('current-menu-item'); 25 } 26 }); 27}); 28</script>
28 29 30</head> 31 32<body class="home"> 33<div id="sidebar"> 34 <header id="masthead" class="site-header" role="banner"> 35 <hgroup> 36 <img class="logo_institution" src="/static/programme/images/logo_palaiseau.png" /> 37 <!-- <img class="logo_institution" src="/static/programme/images/logo_ipp.png" /> --> 38 <div class="info"> 39 <h2 class="site-title"><a href="/about/">Le Seminaire Palaisien</a></h2> 40 <h4 class="site-description">Stat / Math / Info</h4> 41 <h5 class="email"> 42 thomas.moreau [AT] inria.fr <br/> 43 victor.emmanuel.brunel [AT] ensae.fr <br/> 44 tobias.liaudat [AT] cea.fr 45 </h5> 46 </div> 47 </hgroup> 48 49 <nav id="site-navigation" class="main-navigation" role="navigation"> 50 <div class="menu-menu-1-container"><ul id="menu-nav" class="menu-nav"> 51 <li class="nav-menu-item"><a href="/about/">About</a></li> 52 <li class="nav-menu-item"><a href="/talks/">Past Talks</a></li> 53 </ul></div> 54 </nav><!-- #site-navigation --> 55 <div id="social-media"> 56 <img class="logo_institution" src="/static/programme/images/logo_dataia.png" /> 57 <img class="logo_institution" style="padding: 2em 0 1em 0;" src="/static/programme/images/logo_ipp.png" /> 58 <img class="logo_institution" src="/static/programme/images/logo_UPSA.png" /> 59 </div> 60 </header><!-- #masthead --> 61</div> 62</div> 63 64<div id="page"> 65 <div class="page-content"> 66 67 68 69<article class="inner-text"> 70 <header class="entry-header"> 71 <h1 class="entry-title">About the seminar</h1> 72 </header> 73 74 <div class="entry-content"> 75 <p> 76 This seminar aims to increase the links between the different laboratories in Saclay in the field of Applied Maths, Statistics and Machine Learning. The Seminar is organized every first Tuesday of the month with 2 presentations followed by a small refreshment. The localization of the seminar will change to accommodate the different labs. 77 </p> 78 <h3>Organization</h3> 79 <p> 80 Due to access restriction, you need to register for the seminar. A link is provided in the description and should also be sent with the seminar announcement. It will also help us organize for the food quantities. If you think you will come, please register! (even if you are unsure) 81 </p> 82 <p> 83 To not miss the next seminar, please subscribe to the announcement mailing list <a href="https://sympa.inria.fr/sympa/subscribe/palaisien">subscribe/palaisien</a>.<br/> 84 You will receive email on the <code>[email protected]</code> mailing list. <br/> 85 You can also add the calendar from the seminar to your own calendar (see below). 86 </p> 87 88 <h3>Next seminars</h3> 89 <div class="seminar"> 90 <div class="seminar_item"> 91 92 <a class="register" href="/register/51/"> 93 REGISTER 94 </a> 95 96 <span class="date">06 Oct 2026, 12h </span> 97 98 <span class="conference">At <i>Inria Saclay - Amphi Sophie Germain</i></span> 99 100 </div> 101 102 <div class="description"> 103 104 105 <div class="talk"> 106 <div class="talk_item"> 107 108 <a href="https://jualberge.github.io/"> 109 110 <span class="speaker">Julie Alberge</span> 111</a> 112 113 <span>-</span> 114 <span class="title">SurvICL: In-Context Learning for Survival Analysis </span> 115 <span style="width:2em;"></span> 116 117 118 119 120 </div> 121 122 <input type="checkbox" class="btnCtrl" id="talk91"/> 123 <label class="btn display-status" for="talk91"> 124 <i class="fa fa-plus-circle plus"></i> 125 <i class="fa fa-minus-circle minus"></i> 126 </label> 127 <div class="description"> 128 <div class="summary"> 129 Survival analysis aims to predict the time until an event of interest, such as death, disease progression, or equipment failure. Unlike standard regression, it must account for censoring, which occurs when the event has not been observed for some individuals, for example because they leave the study or the study ends. This makes survival prediction particularly challenging, especially on small datasets commonly encountered in clinical research. Recent advances in tabular foundation models... 130 </div> 131 <div class="details"> 132 Survival analysis aims to predict the time until an event of interest, such as death, disease progression, or equipment failure. Unlike standard regression, it must account for censoring, which occurs when the event has not been observed for some individuals, for example because they leave the study or the study ends. This makes survival prediction particularly challenging, especially on small datasets commonly encountered in clinical research.�</br> 133�</br> 134Recent advances in tabular foundation models have demonstrated remarkable performance in classification and regression, often outperforming traditional methods such as gradient-boosted trees. However, their success has not yet translated to survival analysis, where the Cox proportional hazards model, introduced over 50 years ago, remains a strong baseline.�</br> 135�</br> 136In this talk, we introduce SurvICL, a tabular foundation model that predicts individual survival curves through in-context learning, without dataset-specific training or hyperparameter tuning. SurvICL is pretrained entirely on synthetic datasets generated from a flexible proportional-odds prior and learns to predict survival curves using a simple quantile regression loss. We also present our benchmark of 63 survival datasets, the largest to date, and show how SurvICL outperforms 20 competing methods, including classical statistical models, tree-based approaches, deep learning methods, and other foundation models. In particular, SurvICL is the first tabular foundation model to outperform Cox on small survival datasets. 137 </div> 138 </div> 139 </div> 140 141 142 143 <div class="talk"> 144 <div class="talk_item"> 145 146 <a href="https://enzotarta.github.io/"> 147
148 <span class="speaker">Enzo Tartaglione</span> 149</a> 150 151 <span>-</span> 152 <span class="title">Bringing Training to the Edge: Subspace and Sparsity for Efficient Deep Learning </span> 153 <span style="width:2em;"></span> 154 155 156 157 158 </div> 159 160 <input type="checkbox" class="btnCtrl" id="talk87"/> 161 <label class="btn display-status" for="talk87"> 162 <i class="fa fa-plus-circle plus"></i> 163 <i class="fa fa-minus-circle minus"></i> 164 </label> 165 <div class="description"> 166 <div class="summary"> 167 Deep models keep growing, which makes them hard to deploy and even harder to train on edge devices. Classic pruning often fails to deliver real speed-ups. Unstructured sparsity is poorly supported by general-purpose hardware, and on parallel hardware, thinner layers do not shorten the critical path. In this talk, I first present ways to reduce network depth by collapsing layers, removing non-linearities so adjacent layers can be merged (EASIER, TLC, LaCoOT), and to reduce redundant tokens in ... 168 </div> 169 <div class="details"> 170 Deep models keep growing, which makes them hard to deploy and even harder to train on edge devices. Classic pruning often fails to deliver real speed-ups. Unstructured sparsity is poorly supported by general-purpose hardware, and on parallel hardware, thinner layers do not shorten the critical path. In this talk, I first present ways to reduce network depth by collapsing layers, removing non-linearities so adjacent layers can be merged (EASIER, TLC, LaCoOT), and to reduce redundant tokens in multimodal models (FOLDER). I then discuss why pruning at initialization remains hard, and how training itself can be made cheaper by freezing neurons that have reached equilibrium (NEq). Finally, I address memory, the main bottleneck for on-device learning. Because the optimization subspace is stable during fine-tuning, activations and weights can be compressed together through low-rank decomposition. This cuts backpropagation memory and FLOPs by orders of magnitude. I close with open questions on foundation models, continual learning, pruning versus quantization, and links and opportunities to other (more or less) related fields. 171 </div> 172 </div> 173 </div> 174 175 176 </div> 177 </div> 178 179 <div class="seminar"> 180 <div class="seminar_item"> 181 182 <a class="register" href="/register/52/"> 183 REGISTER 184 </a> 185 186 <span class="date">10 Nov 2026, 12h </span> 187 188 <span class="conference">At <i>Inria Saclay - Amphi Sophie Germain</i></span> 189 190 </div> 191 192 <div class="description"> 193 194 </div> 195 </div> 196 197 <br/> 198 <div width=100% style="text-align: left;"> 199 <iframe src="https://calendar.google.com/calendar/embed?height=300&wkst=2&bgcolor=%23ffffff&ctz=Europe%2FParis&src=aWV1ZjdxZmQ2N3QxY2F2MWNsMnBvN3U2ajRAZ3JvdXAuY2FsZW5kYXIuZ29vZ2xlLmNvbQ&color=%23B39DDB&showTitle=1&mode=MONTH&showNav=1&showTabs=0&showPrint=0&showCalendars=0&showTz=0&showDate=1&title=Seminar%20Shared%20Calendar" style="border-width:0" style="width:100%;" width="600" height="300" frameborder="0" scrolling="no"></iframe></div> 200 201 <h3>Scientific Committee</h3> 202 <p> 203 The program and the organization of this seminar is driven by a scie
203ntific committee composed of members of the different laboratories in Saclay. The members of the committee are currently: 204 </p> 205 <ul> 206 <li><a href="https://www.math.u-psud.fr/~arlot/"> 207 Arlot Sylvain</a> (LMO, Université Paris-Sud)</li> 208 <li><a href="http://www.math.uni-potsdam.de/~blanchard/"> 209 Blanchard Gilles</a> (LMO, Université Paris-Sud)</li> 210 <li><a href="https://scholar.google.fr/citations?user=IFkrCzYAAAAJ&hl=en"> 211 Brault Romain</a> (ThereSIS, Thales)</li> 212 <li><a href="https://vebrunel.wordpress.com/"> 213 Brunel Victor-Emmanuel</a> (CREST, ENSAE)</li> 214 <li><a href="https://jchiquet.github.io"> 215 Chiquet Julien</a> (MIA, Agro Paris)</li> 216 <li><a href="http://www.cmap.polytechnique.fr/~aymeric.dieuleveut/"> 217 Dieuleveut Aymeric</a> (CMAP, Ecole polytechnique)</li> 218 <li><a href="https://www.math.u-psud.fr/~goude/about.html"> 219 Goude Yannig</a> (R&D, EDF)</li> 220 <li><a href="http://kalogeratos.com/psite/"> 221 Kalogeratos Argyris</a> (CMLA, ENS Paris-Saclay)</li> 222 <li><a href="http://genome.jouy.inra.fr/~ekuhn/"> 223 Kuhn Estelle</a> (MaIAGE, Inra)</li> 224 <li><a href="https://tommoral.github.io"> 225 Moreau Thomas</a> (Parietal, Inria)</li> 226 <li><a href="https://perso.telecom-paristech.fr/mozharovskyi/"> 227 Mozharovskyi Pavlo</a> (LTCI, Telecom Paris)</li> 228 <li><a href="http://fredericpascal.blogspot.com/"> 229 Pascal Frederic</a> (L2S, CentraleSupelec)</li> 230 <li><a href="https://sites.google.com/site/gquercini/"> 231 Quercini Gianluca</a> (LRI, CentraleSupelec)</li> 232 <li><a href="https://myriamtami.github.io/"> 233 Tami Myriam</a> (MICS, CentraleSupelec)</li> 234 </ul> 235 236 <h3>Funding</h3> 237 <p> 238 This seminar is made possible with financial support of the ENSAE and DataIA. 239 </p> 240 <div> 241 <img class="logo_funding" 242 src="/static/programme/images/logo_dataia_bleu.png" /> 243 <img class="logo_funding" style="height: 15em;" 244 src="/static/programme/images/logo_ensae_black.png" /> 245 </div> 246</div><!-- .entry-content --> 247 248</article><!-- #post --> 249 250 </div> 251</div><!-- #page --> 252 253<div style="display:none"> 254</div> 255 256</body> 257
258<script type="text/javascript"> 259 window.onresize = function(){ 260 var sidebar = document.getElementById("sidebar"); 261 var page = document.getElementById("page"); 262 page.style.setProperty("min-height", sidebar.clientHeight); 263 } 264 window.onresize(); 265</script>
265 266 267 268</html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.