{"id":26798,"date":"2026-03-09T17:27:06","date_gmt":"2026-03-09T16:27:06","guid":{"rendered":"https:\/\/lium.univ-lemans.fr\/?p=26798"},"modified":"2026-03-12T10:24:50","modified_gmt":"2026-03-12T09:24:50","slug":"inlm","status":"publish","type":"post","link":"https:\/\/lium.univ-lemans.fr\/en\/inlm\/","title":{"rendered":"InLM"},"content":{"rendered":"<div class=\"panel-grid\" id=\"pg-26798-0\" ><div class=\"panel-grid-core\"><div class=\"panel-grid-cell\" id=\"pgc-26798-0-0\" ><div class=\"panel-widget-style\" ><h2 style=\"color: #e5442d;\">Interpretable Language Models (InLM)<br\/><\/h2><p><b>Date: <\/b> 01\/2026 - 06\/2027<br\/><b>Funding: <\/b> R\u00e9gion Pays de la Loire<br\/><b>Call: <\/b> \u00c9toiles montantes en Pays de la Loire 2025<br\/><b>URL: <\/b><a href =\"https:\/\/lium.univ-lemans.fr\/en\/inlm\/\">https:\/\/lium.univ-lemans.fr\/en\/inlm\/<\/a><br\/><div class=\"table-responsive\" style=\"margin-left:-20px;\" ><table class=\"table\" style=\"border: 0 solid #ffffff;\"><tr class=\"col-sm-12\" ><td style=\"border: 0;\"><br\/><b>LIUM Participant(s): <\/b><\/td><td style=\"border: 0\"><center><a href=\"https:\/\/cv.hal.science\/nicolas-dugue\" target=\"_blank\" ><img alt=\"User Pic\" src=https:\/\/lium.univ-lemans.fr\/wp-content\/uploads\/2018\/01\/PhotoNicolas.jpg class=\"img-circle img-responsive\" height=\"60\" width=\"60\"><\/a><a href=\"https:\/\/cv.hal.science\/nicolas-dugue\" target=\"_blank\" ><b style=\"color:#e5442d;\"><span style=\"font-size: 8pt;\">Nicolas Dugu\u00e9<\/span><\/b><\/a><\/center><\/td><\/tr><\/table><\/div><\/div><br\/><\/p><\/div><div class=\"panel-widget-style\" ><p>Since the advent of deep learning, the performance of natural language processing (NLP) systems has improved considerably, leading to the emergence of large language models (LLMs). Three factors have contributed to these advances. <\/p>\n<ul>\n<li>The first factor is the development of systems capable of immersing vocabulary in a semantic space, thereby providing rich representations of meaning to machine learning algorithms. <\/li>\n<li>The second factor is the development of architectures adapted to textual data, which is sequential in nature. <\/li>\n<li>Finally, the combined increase in data storage capacity and computing power has made it possible to increase the size of artificial intelligence (AI) system architectures. <\/li>\n<\/ul>\n<p>While the first two points represent undeniable progress, the growth of AI architectures comes with two less positive trade-offs.<br \/>\nThe first is the interpretability of results. The larger and more complex the architectures, the less easy it is to understand the resulting model. The second is energy consumption. Training these systems sometimes requires enormous computing power and time. <\/p>\n<p>With InLM, we want to initiate an innovative response to the issue of interpretability, taking computation times into account. We therefore propose to build and evaluate end-to-end interpretable encoder architectures that are resource-efficient, since their interpretability does not require post-hoc system learning. Interpretability builds trust among end users, particularly in sensitive areas such as justice and medicine, and such encoder architectures can be specialized on hardware that is accessible to all. To achieve this, we are drawing on the interpretable embedding methods developed in the ANR DIGING project, and we will adapt the self-attention paradigm of transformers to ensure interpretability.<\/p><\/div><\/div><\/div><\/div><div class=\"panel-grid\" id=\"pg-26798-1\" ><div class=\"panel-grid-core\"><div class=\"panel-grid-cell\" id=\"pgc-26798-1-0\" >&nbsp;<\/div><div class=\"panel-grid-cell\" id=\"pgc-26798-1-1\" >&nbsp;<\/div><div class=\"panel-grid-cell\" id=\"pgc-26798-1-2\" >&nbsp;<\/div><\/div><\/div><div class=\"panel-grid\" id=\"pg-26798-2\" ><div class=\"panel-grid-core\"><div class=\"panel-grid-cell\" id=\"pgc-26798-2-0\" >&nbsp;<\/div><\/div><\/div><div class=\"panel-grid\" id=\"pg-26798-3\" ><div class=\"panel-grid-core\"><div class=\"panel-grid-cell\" id=\"pgc-26798-3-0\" >&nbsp;<\/div><div class=\"panel-grid-cell\" id=\"pgc-26798-3-1\" ><div class=\"panel-widget-style\" ><p><img src=\"https:\/\/lium.univ-lemans.fr\/wp-content\/uploads\/2026\/03\/logo-Pays-de-la-loire.png\" alt=\"\" \/ ><\/p><\/div><\/div><div class=\"panel-grid-cell\" id=\"pgc-26798-3-2\" >&nbsp;<\/div><\/div><\/div>","protected":false},"excerpt":{"rendered":"<p>Interpretable Language Models (InLM)Date: 01\/2026 &#8211; 06\/2027Funding: R\u00e9gion Pays de la LoireCall: \u00c9toiles montantes en Pays de la Loire 2025URL: https:\/\/lium.univ-lemans.fr\/en\/inlm\/LIUM Participant(s): Nicolas Dugu\u00e9Since the advent of deep learning, the performance of natural language processing (NLP) systems has improved considerably, leading to the emergence of large language models (LLMs). Three factors have contributed to these [&hellip;]<\/p>\n<p class=\"more-link style2\"><a href=\"https:\/\/lium.univ-lemans.fr\/en\/inlm\/\"  class=\"themebutton\"><span class=\"more-text\">READ MORE<\/span><span class=\"more-icon\"><i class=\"fa fa-angle-right fa-lg\"><\/i><\/span><\/a><\/p>\n","protected":false},"author":14,"featured_media":13261,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[45],"tags":[49],"acf":[],"_links":{"self":[{"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/posts\/26798"}],"collection":[{"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/users\/14"}],"replies":[{"embeddable":true,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/comments?post=26798"}],"version-history":[{"count":3,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/posts\/26798\/revisions"}],"predecessor-version":[{"id":26811,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/posts\/26798\/revisions\/26811"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/media\/13261"}],"wp:attachment":[{"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/media?parent=26798"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/categories?post=26798"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lium.univ-lemans.fr\/en\/wp-json\/wp\/v2\/tags?post=26798"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}