Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Le groupement artisanal de l’Antiquité tardive de Las Cravieros à Fanjeaux (Aude) : lieu de production de poteries, de matériaux de construction et du travail du fer – troisième rapport de la 2e triennale (2022-2024), 2024. Rapport de fouilles
No full textMicroarchitectural signals analysis platform for the implementation of Hardware Security Counters
No full textInternational audienceDetecting malicious software or hardware behavior during the operation of a computer system requires observables from one or more abstraction layers of the system. This abstraction, however, tends to limit the ability to detect behavioral deviations, especially for attack classes that exploit vulnerabilities very close to the target hardware. Conversely, too low a level of abstraction tends to significantly increase the complexity of the system model, and therefore poses a number of difficulties for the extraction and selection of relevant observables for a given class of attack.Hardware performance counters in particular have been used as an indirect means of observing micro-architecture behavior and detecting software attempting to exploit hardware vulnerabilities. In order to improve the various detection methods, we propose the construction of hardware metrics designed from the outset for security, by studying the correlation between signals from the micro-architecture and the various classes of attack in the literature, targeting both conventional IT and industrial OT systems. By extension, this work aims to detect attacks originating from hardware Trojans, the latter having the effect of changing the behavior of a given micro-architecture
Plateforme d'analyse de signaux micro-architecturaux pour la mise en place de compteurs matériels de sécurité
No full textInternational audienceLa détection de comportements malveillants logiciels ou matériels pendant le fonctionnement d'un système informatique demande la mise en place d'observables provenant d'une ou plusieurs couches d'abstraction de ce dernier. Cette abstraction cependant tend à limiter la capacité à détecter des déviations de comportement, surtout pour des classes d'attaques qui exploitent des vulnérabilités très proches du matériel cible. A contrario, un niveau d'abstraction trop faible tend à faire croître significativement la complexité du modèle du système et donc pose un certain nombre de difficultés pour l'extraction et la sélection des observables pertinents pour une classe d'attaque donnée.Les compteurs de performance des processeurs ont notamment été utilisés comme moyen indirecte d'observer le comportement de la micro-architecture et détecter des logiciels tentant d'exploiter des vulnérabilités matérielles. Afin d'améliorer les différentes méthodes de détection, nous proposons la construction de métriques matérielles pensées dès la conception pour la sécurité en étudiant la corrélation entre les signaux provenant de la micro-architecture et les différentes classes d'attaque de la littérature ciblant à la fois des systèmes IT classiques et OT industriels.Par extension, ces travaux ont pour ambition de détecter des attaques provenant de chevaux de Troie matériels, ces derniers ayant pour effet de changer le comportement d'une micro-architecture donnée
First Order Model-Based RL through Decoupled Backpropagation
No full textInternational audienceThere is growing interest in reinforcement learning (RL) methods that leverage the simulator's derivatives to improve learning efficiency. While early gradient-based approaches have demonstrated superior performance compared to derivative-free methods, accessing simulator gradients is often impractical due to their implementation cost or unavailability. Model-based RL (MBRL) can approximate these gradients via learned dynamics models, but the solver efficiency suffers from compounding prediction errors during training rollouts, which can degrade policy performance. We propose an approach that decouples trajectory generation from gradient computation: trajectories are unrolled using a simulator, while gradients are computed via backpropagation through a learned differentiable model of the simulator. This hybrid design enables efficient and consistent first-order policy optimization, even when simulator gradients are unavailable, as well as learning a critic from simulation rollouts, which is more accurate. Our method achieves the sample efficiency and speed of specialized optimizers such as SHAC, while maintaining the generality of standard approaches like PPO and avoiding ill behaviors observed in other first-order MBRL methods. We empirically validate our algorithm on benchmark control tasks and demonstrate its effectiveness on a real Go2 quadruped robot, across both quadrupedal and bipedal locomotion tasks
Problèmes de stabilité pour les automates cellulaires probabilistes
No full textCellular automata can be defined as discrete dynamical systems that synchronously and uniformly update symbols from a finite alphabet placed on a network based on the neighboring symbols. In this thesis, we look at random perturbation of such systems: each update, independently, has a probability to make a mistake. When this perturbation becomes negligible and the probability of error tends to 0, the accumulation set of invariant measures of the probabilistic cellular automaton associated is called the set of 0-noise limit measures. We first look at the 0-noise limit measures of some usual classes of cellular automata. We also show that the problem of deciding if a probability measure is a 0-noise limit measure of a given cellular automaton or not is actually undecidable. We then compute several topological and computational constraints on the theoretical set of 0-noise limit measures. When this set of measures is uniformly approached, it is then necessarily a connected [dollar]Pi_{2}[dollar]-computable compact set. This is actually a characterization, as we are able to construct a cellular automaton realizing it when perturbed by a standard noise. Following that, we also describe a cellular automaton whose perturbation by a standard noise realizes at least two phase transitions for its ergodicity. More precisely, the associated probabilistic cellular automaton is ergodic when the noise level is small enough or is large enough, but not ergodic for a given level in between. A cellular automaton perturbed by a noise with positive rates is generally ergodic: starting from any configuration, the system converges to a unique measure that only depends on the noise. There exists only one example of a 1-dimensional cellular automaton whose perturbation is not ergodic, which is very complex and true only for a small perturbation. We use it in order to describe a cellular automaton whose perturbation by a standard noise realizes at least two phase transitions for its ergodicity. More precisely, the associated probabilistic cellular automaton is ergodic when the noise level is small enough or is large enough, but not ergodic for a given level in between. Finally, we generalize the previous problems to the class of continuous functions on the set of configurations on itself. We show that one can construct such a continuous function whose probabilistic perturbation is only ergodic for an error level in a countable intersection of open sets on the unit interval.Les automates cellulaires peuvent être définis comme des systèmes dynamiques discrets qui actualisent des symboles d'un alphabet fini placés sur un réseau en fonction des symboles voisins de manière uniforme et synchrone. On s'intéresse dans cette thèse à des perturbations aléatoires de ces systèmes : chaque actualisation, de manière indépendante, a une probabilité d'être erronée. Lorsque cette perturbation devient négligeable et que la probabilité d'erreur tend vers 0, on appelle les points d'accumulation des mesures invariantes de l'automate cellulaire probabiliste associé les mesures 0-limites. Dans un premier temps, on s'intéresse aux mesures 0-limites de certaines classes usuelles d'automates cellulaires. On montre cependant que le problème de savoir si une mesure de probabilité est une mesure 0-limite pour un automate cellulaire donné perturbé par un bruit standard est indécidable. Dans un second temps, on précise quelles sont les contraintes topologiques et de calculabilité qui s'appliquent sur les ensembles de mesures 0-limites théoriquement possibles. Lorsque l'ensemble en question est uniformément approché, c'est alors nécessairement un compact connexe [dollar]Pi_{2}[dollar]-calculable. C'est en réalité une caractérisation, puisqu'on est capable de construire un automate cellulaire qui le réalise sous une perturbation standard. Un automate cellulaire perturbé par un bruit à taux positifs est généralement ergodique: partant de n'importe quelle configuration, le système converge vers une unique mesure qui dépend uniquement du bruit. Il n'existe qu'un seul exemple d'automate cellulaire unidimensionnel dont la perturbation n'est pas ergodique, qui est très complexe et vrai uniquement pour une perturbation très faible. On l'utilise pour décrire dans un troisième temps un automate cellulaire dont la perturbation standard réalise deux changements de phase pour l'ergodicité. Plus précisément, l'automate cellulaire probabiliste associé est ergodique pour une perturbation assez faible et pour une perturbation assez forte, mais non-ergodique pour un taux d'erreur fixé entre les deux. Enfin, on considère les questions précédentes appliquées aux fonctions continues d'un ensemble de Cantor sur lui-même. On montre ainsi qu'il est possible de construire une fonction continue dont la perturbation aléatoire est ergodique uniquement pour un taux d'erreurs appartenant à une intersection dénombrable d'ouverts de l'intervalle arbitraire
Le parc de Canudos, un lieu de mémoire au cœur de la « forêt blanche » (État de Bahia, Brésil)
No full texthttps://senforester.hypotheses.org
Measuring the Transparency of the Derived Portuguese Lexicon: Integrating Computational and Behavioral Data: Integrating Computational and Behavioral Data
No full textInternational audienceThe main goal of our work is to investigate the effect of morphological transparency on the accessibility of the derived lexicon in Portuguese. In particular, we focus on deverbal nouns derived by suffixation and denoting agents (investidor ‘investor’) or actions (instalação ‘installation’). We compare transparency measures obtained from corpus data with the results of a behavioral experiment conducted with native speakers of Brazilian Portuguese. Through a corpus analysis, we identify deverbal nouns in Portuguese that are likely to be activated more quickly and categorize them into three levels of transparency. Words from each of these three categories are then used as material for a lexical decision task. The aim is to assess the relevance of corpus-based measures in the context of a behavioral experiment.As a first step, 4,016 deverbal nouns were extracted from the Brazilian Portuguese Lexicon corpus (Estivalet & Meunier, 2017). The nouns selected contained one of the productive suffixes in Portuguese that form action (-agem, -ão, -ança/-ância, -mento) or agent nouns (-ante/-ente, -dor). Each of the nouns in the database was assigned a transparency score, calculated by combining three metrics which are generally considered as good indicators of transparency in the literature: a) the Levenshtein distance (Levenshtein 1966) between the orthographic forms of the verbal and nominal stems, normalized according to the length of the stem (cf. Strnadová, 2014: 112-113); b) the ratio between the frequency of the base and the frequency of the derived noun (cf. Hay, 2001); c) the size of the morphological family (i.e. the number of derivatives in the corpus containing the same suffix, cf. Moscoso del Prado Martín et al., 2004; Racz et al., 2016). From the combination of these three measures, we obtained a ranking of the words in our database based on their transparency score. Words were then divided into three numerically homogeneous groups (1336 each) corresponding to high (e.g., executante ‘executor’), medium (e.g., comentário ‘comment’), and low (e.g., vigilância ‘vigilance’) transparency.For our behavioural experiment, we realized a lexical decision task on 31 native speakers of Portuguese, measuring accuracy rates and reaction times. Our sample included a majority of speakers of Brazilian Portuguese, aged between 19 and 56 years (M = 36.9 years), with diverse socio-demographic profiles. The materials used were selected from our database of deverbal nouns, controlled for frequency and length. In total, the experimental materials consisted of 150 words (50 for each transparency class) and 100 pseudowords. A logistic regression model shows that accuracy in recognizing less transparent words decreased by 40% compared to more or moderately transparent words. Concerning reaction time, a mixed linear regression model shows a statistically significant difference (p = 0.00508) between opaque and transparent words. As transparency decreases, reaction time increases
Effective Data Generation and Feature Selection in Learning for Planning
No full textInternational audiencePrevious studies have shown that leveraging data beyond optimal training plans improves the learning of search guidance for planning. Specifically, state ranking information can be extracted from states on optimal plan traces and their siblings. In this paper, we generalise this approach by extracting additional rankings from the A⋆ search tree for generating optimal training plans. As in the previous approach, we incur no additional search effort and negligible computational overhead for data extraction. However, extracting more data in this way may introduce many redundant features and states which slows down training. We formalise the problem of sound, redundant feature pruning and show that it is NP-complete to solve. Furthermore, we introduce several algorithms and approximations for redundant feature pruning. Experiments show that rankings learned by extracting more data from search trees for generating optimal training plans improve planner coverage. However, pairing with unsound pruning methods often results in diminishing performance, while our sound feature pruning methods provide consistent improvements across tested domains
Engineering Ni-Silicide Nanocontacts for 3D Silicon Devices via Geometrical Confinement Control
No full textInternational audienceNanoscale Ni-silicide alloys are critical components for future generations of 3D electronic devices based on active Si nanostructures, with applications in nanoelectronics, energy conversion, and sensing. This study investigates how geometrical confinement in such nanostructures influences diffusion-driven silicidation, ultimately determining the alloy formation sequence, phase composition, and volumetric expansion. The silicidation of controlled Ni volumes is investigated on vertical silicon nanowires (NW) and nanosheets (NS) under various annealing conditions. The silicide phases and interface morphologies are characterized using high-resolution (scanning) transmission electron microscopy (HR-TEM, HR-STEM), energy-dispersive X-ray spectroscopy (EDX), and four-dimensional scanning transmission electron microscopy (4D-STEM) for nanoscale Ni−Si phase mapping. Under conditions of strong geometric confinement, NiSi2 is observed to form with faceted, prism-like morphologies aligned with Si (111) planes, features not typically present in planar or bulk samples. This anisotropic growth is associated with preferential Ni diffusion along nanostructure surfaces and limited Si counterdiffusion through the silicide. The resulting NiSi2 interfaces are structurally distinct and may contribute to reduced contact resistance in both p-type and n-type silicon nanostructures, supporting their integration in 3D device architectures
Pensée visuelle et imagerie mentale dans l’autisme : revue de la littérature
No full textInternational audienceIntroduction. Le trouble du spectre de l’autisme se caractérise par des déficits persistants dans la communication et les interactions sociales, ainsi que par des particularités sensorielles et perceptives. En lien avec ces spécificités, les personnes autistes sont fréquemment décrites comme ayant un mode de pensée plus visuel. Toutefois, bien que l’hypothèse d’une « pensée en images » soit largement relayée dans la littérature, elle repose sur un nombre limité de preuves expérimentales. Objectif. L’objectif de cette revue narrative de la littérature est d’évaluer l’hypothèse d’une pensée plus visuelle dans l’autisme à travers l’analyse de différentes manifestations de cette pensée telles que l’imagerie mentale, la synesthésie et le style cognitif visuel. Méthodologie. Les sources ont été sélectionnées à partir d’une recherche documentaire, menée entre 1990 et 2023, à l’aide de moteurs de recherche spécialisés, à partir des mots-clés : autisme, pensée visuelle, imagerie mentale, synesthésie et style cognitif. Résultats. Les résultats de cette revue indiquent que les personnes autistes présentent des capacités supérieures de manipulation des images mentales, s’appuyant sur un réseau cérébral distinct de celui mobilisé par les personnes neurotypiques. Par ailleurs, le phénomène de synesthésie semble plus fréquent chez les personnes autistes, et l’utilisation de stratégies visuelles apparaît privilégiée dans leur fonctionnement quotidien. Conclusion. Ces données soutiennent l’hypothèse d’un style cognitif à dominante visuelle chez les personnes autistes. Néanmoins, les capacités d’imagerie mentale dans l’autisme méritent d’être étudiées plus en profondeur, en tenant compte de l’importante hétérogénéité des profils cognitifs. De plus, l’impact fonctionnel de cette imagerie mentale développée demeure à explorer