Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Computing Approximate Nash Equilibria for Integer Programming Games
No full textInternational audienceWe propose a framework to compute approximate Nash equilibria in integer programming games with nonlinear payoffs, i.e., simultaneous and non-cooperative games where each player solves a parametrized mixed-integer nonlinear program. We prove that using absolute approximations of the players' objective functions and then computing its Nash equilibria is equivalent to computing approximate Nash equilibria where the approximation factor is doubled. In practice, we propose an algorithm to approximate the players' objective functions via piecewise linear approximations. Our numerical experiments on a cybersecurity investment game show the computational effectiveness of our approach
Into the wild. Repenser les rencontres protohistoriques entre les sociétés humaines et leur environnement. XIe Rencontres doctorales de l'école européenne de Protohistoire de Bibracte, Bibracte, 12-16 mars 2025
No full textInternational audienc
Further Bounding the Kreuzer-Skarke Landscape
No full textInternational audienceBatyrev's construction provides a map from fine, regular, star triangulations (FRSTs) of 4D reflexive polytopes to smooth Calabi-Yau threefolds (CYs). We prove that there are at most diffeomorphism classes of CYs produced in this manner, improving arXiv:2008.01730's upper bound of . To show this, we make use of the fact that any two FRSTs with the same 2-face restrictions give rise to diffeomorphic CYs and bound the number of such '2-face equivalence classes' for all polytopes with Hodge number . We also put a lower bound of on the number of 2-face equivalence classes, but emphasize that this is not a lower bound on the number of diffeomorphism classes of CYs, as distinct 2-face equivalence classes may give rise to diffeomorphic threefolds
Transformer les systèmes d’élevage par le jeu ?: Retour sur le processus de conception d’un jeu sérieux destiné à accompagner la transition des systèmes cunicoles
No full textInternational audienceInsta'Lap is a serious game that aims to support the transition of rabbit farming systems. This is a simulation game that enables players to (re)design a rabbit farming system that is desirable for them and acceptable to society. The course of a game includes a phase in which players identify their aspirations and farming practices using a selection of cards, a phase of collective reflexivity, and a multi-criteria assessment of sustainability of the systems designed. At the crossroads of social sciences (sociology and education) and animal sciences, the game design process is examined through the prism of the difficulties and frictions encountered in the co-design committee. Insta’Lap is considered as a “frontier object” that promotes dialogue between social worlds of rabbit farming gathered within the co-design committee. The discussion questions the playfulness and the gameplay of Insta’Lap and the transformative potential of play.Insta’Lap est un jeu sérieux qui vise à accompagner la transition des systèmes d’élevage cunicole. Il s’agit d’un jeu de simulation permettant aux joueur·euses de concevoir un système d’élevage de lapins désirable pour eux et acceptable pour la société. Le déroulement d’une partie inclut une phase d’identification d’aspirations et de pratiques d’élevage grâce à des cartes au choix, une phase de réflexivité collective, et une phase d’évaluation multicritère de la durabilité des systèmes conçus. Au croisement des sciences sociales (sociologie et sciences de l’éducation) et de la zootechnie, le processus de conception du jeu est examiné au prisme des difficultés et frictions rencontrées dans le comité de co-conception. Insta’Lap est considéré comme un « objet-frontière » favorisant le dialogue entre les mondes sociaux cunicoles réunis au sein de ce comité. La discussion s’interroge sur la ludicité, la jouabilité d’Insta’lap et le potentiel transformatif du jeu
Fêtes musicales de 1835 à Toulouse: dossier de presse, chronologie et programmes des 3 événements, Dezède [en ligne], https://dezede.org/dossiers/id/964/
No full textdossier de presse, chronologie et programmes des 3 événements, Dezède [en ligne], https://dezede.org/dossiers/id/964
Réseaux de neurones graphiques explicables pour la détection et le diagnostic d'anomalies dans les séries temporelles multivariées
No full textInternational audienceIndustrial systems consist of multiple interconnected sensors generating multivariate time series (MTS), where anomaly detection is critical for ensuring reliability and safety. While recent Graph Neural Network (GNN)–based approaches effectively detect when anomalies occur, they often fail to explain which sensor measurements are truly symptomatic and why anomalies emerge.We propose an explainability framework that combines attention-based GNN anomaly detection with consistency-based model-based diagnosis (MBD). Under the hypothesis that anomalies arise not from faulty sensors themselves but from altered inter-sensor relationships, we analyze attention changes between a normal model and a fault-adapted model to identify influential relational components responsible for system failures. By integrating attention variation analysis with conflict-set generation and minimal hitting set computation, our method produces ranked candidate diagnoses corresponding to faulty system influences
Problèmes ouverts en géométrie symplectique quantitative et étude des billards
No full textThis document collects contributions to the Open Problem List in Billiards and Quantitative Symplectic Geometry, compiled following discussions during the workshop "Billiards and quantitative symplectic geometry" that took place at the University of Heidelberg on July 14-18, 2025
Smart sampling for optimized materials: A Gaussian process framework for region-of-interest active learning
No full textInternational audienceDiscovering materials with properties that meet user-defined performances remains a critical yet resource-intensive challenge in materials engineering, often requiring costly experiments or simulations. This paper introduces a data-efficient Gaussian Process (GP) active learning framework designed to map multidimensional design spaces and efficiently identify subsets of the output space that represent the optimal or acceptable range for a specific application, defined as Regions of Interest (RoIs). Three novel acquisition function families were proposed: (i) Probability-Entropy acquisitions, balancing exploration and exploitation without additional tuning; (ii) Dual Level-Set Gaussian Bound, extending classical straddle heuristics to multiple thresholds; and (iii) Expected Mahalanobis (ExM), generalizing boundary-focused sampling to arbitrary multivariate target distributions. Benchmark evaluations on synthetic functions and a Al/CuO thermite powder mixture design case demonstrate that the ExM acquisition significantly outperforms existing methods, achieving faster RoI discovery, lower uncertainty, and more complete mapping with minimal iterations. We found that ExM removes the need to tune an explicit exploration parameter. It only requires choosing a target distribution within the RoI, for which we provide simple default options (Uniform or truncated Normal) along with practical guidelines. Applied to Al/CuO thermite formulation for welding applications, ExM efficiently identifies diverse optimized compositions within the target RoI, minimizing redundancy and experimental cost. Overall, ExM provides a scalable and flexible RoI-targeted active learning approach that removes exploration-weight tuning for materials design and beyond
TRACT : A Transformer and Statistical Framework for Anomaly Detection in Multivariate Non-stationary Time Series
No full textInternational audiencePhysiological and inertial signals analysis supports numerous healthcare applications, including disease detection, rehabilitation, and treatment. Advancements in signal processing enable the representation of most stationary and non-stationary phenomena using mathematical expressions. These representations provide valuable insights and help in identifying distinctive patterns of interest. In this paper, we propose TRACT, a deep learning and statistical framework designed to detect anomalies in non-stationary environments. It comprises two main components, a transformer-based reconstruction model that captures signal patterns through multi-resolution attention, extending the standard attention mechanism in transformer architecture. During inference, reconstruction errors are computed by comparing observed signals with their reconstructed versions. Statistical modeling is applied to these errors, with parameters estimated directly from the data. TRACT adapts to varying data rates across datasets without imposing strict distribution assumptions, resulting in enhanced robustness and accuracy in anomaly detection for multivariate non-stationary time series. We evaluate TRACT on 12 real-world multivariate time series datasets from diverse domains, demonstrating its performance in anomaly detection tasks with various constraints and its ability to provide early warnings for anomalous events
Diffusion-Based Fourier Domain Deconvolution with Application to Ultrasound Image Restoration
No full textInternational audienceDeconvolution remains an ongoing challenge in ultrasound imaging. Model-based techniques require extensive parameter tuning and are computationally demanding, whereas convolutional neural networks struggle due to the scarcity of labeled US data. To overcome these challenges, we propose a prior-driven, weakly supervised, diffusion-based deconvolution method. By operating in the Fourier domain, our approach improves image quality while maintaining computational efficiency. Experimental results on in silico and in vivo datasets demonstrate the effectiveness of our method compared to benchmarks