Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
ReToP: Learning to Rewrite Electronic Health Records for Clinical Prediction
No full textAccepted by WSDM 2026International audienceElectronic Health Records (EHRs) provide crucial information for clinical decision-making. However, their high-dimensionality, heterogeneity, and sparsity make clinical prediction challenging. Large Language Models (LLMs) allowed progress towards addressing this challenge by leveraging parametric medical knowledge to enhance EHR data for clinical prediction tasks.Despite the significant achievements made so far, most of the existing approaches are fundamentally task-agnostic in the sense that they deploy LLMs as EHR encoders or EHR completion modules without fully integrating signals from the prediction tasks. This naturally hinders task performance accuracy. In this work, we propose Rewrite-To-Predict (ReToP), an LLM-based framework that addresses this limitation through an end-to-end training of an EHR rewriter and a clinical predictor. To cope with the lack of EHR rewrite training data, we generate synthetic pseudo-labels using clinical-driven feature selection strategies to create diverse patient rewrites for fine-tuning the EHR rewriter. ReToP aligns the rewriter with prediction objectives using a novel Classifier Supervised Contribution (CSC) score that enables the EHR rewriter to generate clinically relevant rewrites that directly enhance prediction. Our ReToP framework surpasses strong baseline models across three clinical tasks on MIMIC-IV. Moreover, the analysis of ReToP shows its generalizability to unseen datasets and tasks with minimal fine-tuning while preserving faithful rewrites and emphasizing task-relevant predictive features
Born to be alive. Isotopic evidence of investment in the survival of immature individuals in the Neolithic period, ecological influences and the construction of social identities
No full textInternational audienc
Long-time behaviour of a multidimensional age-dependent branching process with a singular jump kernel modelling telomere shortening
No full textInternational audienceIn this article, we investigate the ergodic behaviour of a multidimensional age-dependent branching process with a singular jump kernel, motivated by studying the phenomenon of telomere shortening in cell populations. Our model tracks individuals evolving within a continuous-time framework indexed by a binary tree, characterised by age and a multidimensional trait. Branching events occur with rates dependent on age, where offspring inherit traits from their parent with random increase or decrease in some coordinates, while the most of them are left unchanged. Exponential ergodicity is obtained at the cost of an exponential normalisation, despite the fact that we have an unbounded age-dependent birth rate that may depend on the multidimensional trait, and a non-compact transition kernel. These two difficulties are respectively treated by stochastically comparing our model to Bellman-Harris processes, and by using a weak form of a Harnack inequality. We conclude this study by giving examples where the assumptions of our main result are verified
Biologie et sciences humaines: Le point de vue d’un biologiste de l’évolution sur le livre de Bernard Lahire
No full textInternational audienceBernard Lahire’s book, Les Structures fondamentales des sociétés humaines (2023), is a magnificent plea for transdisciplinarity, an approach that I find absolutely essential to enable all sciences to continue to advance our understanding of the universe around us. That being said, I would like to share a few thoughts that came to mind while reading this magnificent work, in relation to my own approach to synthesis in the field of biology. The first thought I would like to address here is in support of the transdisciplinary approach adopted by Bernard Lahire. Next, I offer two thoughts on the importance of general interdisciplinary laws. My fourth thought follows on from the previous ones and concerns the necessity of developing a common vocabulary to promote synthesis between disciplines. Finally, I make the connection with an important topic in the humanities: the origin of inequalities, a subject that Bernard Lahire naturally addresses in his book. My ultimate goal is to remind readers how closely biology and the humanities are linked, in that they both deal with the understanding of living things, which leads them to share many concepts and principles.À mes yeux, le livre de Bernard Lahire, Les Structures fondamentales des sociétés humaines (2023), est un magnifique plaidoyer en faveur de la transdisciplinarité, démarche que je trouve absolument essentielle pour permettre à toutes les sciences, quelles qu’elles soient, de continuer à nous faire progresser dans la compréhension de l’univers qui nous entoure. Cela étant dit, je partage ici quelques réflexions qui m’ont traversé l’esprit en lisant ce magnifique ouvrage, en lien avec ma propre démarche de synthèse dans le domaine de la biologie. La première réflexion que j’aborde ici vise à soutenir la démarche transdisciplinaire adoptée par Bernard Lahire. Ensuite, je propose deux réflexions concernant l’importance des lois générales interdisciplinaires. Ma quatrième réflexion est en continuité avec les précédentes, et concerne le travail nécessaire sur un vocabulaire commun pour favoriser la synthèse entre les disciplines. Enfin, je fais le lien avec un sujet important des sciences humaines : l’origine des inégalités, sujet bien entendu abordé par Bernard Lahire dans son livre. Mon objectif ultime est de rappeler à quel point la biologie et les sciences humaines sont liées dans la mesure où elles portent toutes les deux sur la compréhension du vivant ce qui les conduit à partager de nombreux concepts et principes
Pollution of the scientific literature - Examples and prevention
No full textDoctoralLe mur des connaissances érigé par les générations de chercheurs doit être le plus robuste possible. Or, certaines briques empilées récemment sont friables et menacent l’édifice : même les maisons d’édition les plus réputées publient parfois des articles non fiables. Cette conférence dresse un panorama des méconduites contemporaines qui polluent la littérature scientifique (phrases torturées, paper mills, cartels de citations, lignées cellulaires imaginaires, …), tout en exposant des contre-feux sous la forme d’actions concrètes à intégrer dans nos pratiques pour détecter les problèmes.Présentée le 13/01/26 pour un webinaire organisé par le Service d'Appui à la Recherche et à l'Information Scientifique du CEA Paris-Saclay
Les cirques et leurs spectacles dans les Gaules : état de la question
No full textInternational audienc
Locomotion Mode Transitions: Tackling System- and User-Specific Variability in Lower-Limb Exoskeletons
No full textInternational audienceAccurate detection of locomotion transitions, such as walk to sit, walk to stair ascent, and descent, is crucial to effectively control robotic assistive devices, such as lowerlimb exoskeletons, as each locomotion mode requires specific assistance. Variability in collected sensor data introduced by useror system-specific characteristics makes it challenging to maintain high transition detection accuracy while avoiding latency using non-adaptive classification models. In this study, we identified key factors influencing transition detection performance, including variations in user behavior, and different mechanical designs of the exoskeletons. To boost the transition detection accuracy, we introduced two methods for adapting a finite-state machine classifier to system-and user-specific variability: a Statistics-Based approach and Bayesian Optimization. Our experimental results demonstrate that both methods remarkably improve transition detection accuracy across diverse users, achieving up to an 80% increase in certain scenarios compared to the non-personalized threshold method. These findings emphasize the importance of personalization in adaptive control systems, underscoring the potential for enhanced user experience and effectiveness in assistive devices. By incorporating subjectand system-specific data into the model training process, our approach offers a precise and reliable solution for detecting locomotion transitions, catering to individual user needs, and ultimately improving the performance of assistive devices
FALCON: Unfolded Variational Model for Blind Deconvolution and Segmentation in 3d Dental Imaging
No full textInternational audienceThis paper proposes an unfolded variational network for joint 3D blind deconvolution and segmentation, derived from a Mumford-Shah (MS) estimation. The approach unrolls a Split Bregman optimization into a fixed number of learnable iterations, effectively combining model-driven regularization with data-driven parameter adaptation. When applied to Cone Beam Computed Tomography (CBCT) volumes, the approach restores fine anatomical details approaching micro-CT (µ-CT) quality while maintaining its interpretability through explicit variational terms. A subsequent clustering-based segmentation further refines the reconstructed structures, highlighting the potential of hybrid unfolding schemes to integrate physics-based priors and deep learning (DL) representations in 3D medical imaging