Archive ouverte HAL-LAAS
Not a member yet
12189 research outputs found
Sort by
Pointer-Based Deep Reinforcement Learning Agent for Network Slice Placement in Large Scale Networks
No full textInternational audienceDeep Reinforcement Learning (DRL) has proven effective across numerous fields, making it a promising approach for optimizing network slice placement in 5G mobile networks. Many studies have highlighted the advantages of DRL agents in autonomously and efficiently placing network slices, aligning well with the vision of major operators and organizations for next-generation networks. However, the challenge of stable convergence in DRL agents, particularly when their action space becomes large, has limited their application to smalland medium-sized infrastructures. In this paper, we address this limitation by introducing P-DDQN, a novel pointer-based DRL algorithm that uses a fixed set of navigation actions to enable spatial exploration of the infrastructure. P-DDQN is capable of exploring very large-scale networks while maintaining a constant number of actions. We evaluated it on small, medium, and large infrastructures, and simulation results demonstrate that it outperforms traditional DRL agents, such as Double Deep Q-Network (DDQN) and Deep Q-Network (DQN).</div
Taming the Triangle: On the Interplays Between Fairness, Interpretability, and Privacy in Machine Learning
No full textInternational audienceMachine learning techniques are increasingly used for high‐stakes decision‐making, such as college admissions, loan attribution, or recidivism prediction. Thus, it is crucial to ensure that the models learnt can be audited or understood by human users, do not create or reproduce discrimination or bias and do not leak sensitive information regarding their training data. Indeed, interpretability, fairness, and privacy are key requirements for the development of responsible machine learning, and all three have been studied extensively during the last decade. However, they were mainly considered in isolation, while in practice they interplay with each other, either positively or negatively. In this survey paper, we review the literature on the interactions between these three desiderata. More precisely, for each pairwise interaction, we summarize the identified synergies and tensions. These findings highlight several fundamental theoretical and empirical conflicts, while also demonstrating that jointly considering these different requirements is challenging when one aims at preserving a high level of utility. To solve this issue, we also discuss possible conciliation mechanisms, showing that a careful design can enable to successfully handle these different concerns in practice
Développement d'un système interférométrique pour la mesure de surpression aérienne par rétro-injection optique
No full textThe objective of this thesis was to develop an air overpressure sensor using optical feedback in a laser diode. In detonics experiments, the air overpressure sensors used by CEA Gramat are mostly piezoelectric. The purpose was to create an optical sensor that overcomes the intrinsic limitations of piezoelectric sensors, particularly in terms of bandwidth and electromagnetic disturbances. As part of an initial study conducted at the LICUR research laboratory—a collaboration between LAAS-CNRS and CEA Gramat—the feasibility of a sensor based on optical feedback for shock wave characterization was evaluated. The approach relies on a detailed understanding of the acousto-optic effect under very high pressures and the thermodynamic effects induced by the shock wave. The sensor generates interferometric fringes, linked to optical feedback, that are proportional to the refractive index variations along the laser measurement beam. This thesis describes the complete design of the sensing system. It involved simultaneous work on signal processing development and the progressive validation of the optical feedback sensor under both laboratory and real conditions. The first tests demonstrated both the strong potential of this technology and the inherent challenges of a sensor based on this principle, particularly the detection of high-frequency interference fringes and the decorrelation of pressure measurements from refractive index changes while accounting for thermodynamic effects. These tests were carried out using laboratory setups: a miniature chamber at LAAS and a dynamic shock tube at CEA Gramat. They validated a first version of the acousto-optic model (valid up to 20 bar) and allowed comparison of the derived pressure measurements with time-resolved data obtained using commercial piezoelectric sensors. The experiments also showed that the shape of the shock front influences the optical measurement. Indeed, the optical sensor integrates the optical index along the laser beam, making it necessary to study the influence of the curvature radius to determine the maximum pressure of the shock front. The signal processing aspect of the sensor is also a key point of this thesis. The implementation of a more robust algorithm than previous versions, based on wavelet processing, proved essential for fringe detection. The developed optical sensor was also deployed under real conditions at low levels in a shock tube of about ten centimeters in diameter, and during open-field detonation experiments with pressure levels around 5 bar. The final tests demonstrated the optical sensor's capability to measure the rising front of the shock wave, as well as the robustness of its opto-mechanical design with respect to pressure levels, fireball effects, bright flash, and extreme temperatures during the experiments (-10°C to +40°C). Finally, a comparison was made between the optical feedback approach and that of fibered Michelson interferometry, as both methods are theoretically capable of measuring variations in the optical index due to a shock wave. It was shown that the Michelson interferometer could measure the refractive index with higher resolution (~10⁻⁸) but was more sensitive to laser beam alignment adjustments than the optical feedback sensor, which is currently limited to a single interference fringe in terms of resolution (~10⁻⁵). However, the optical feedback sensor is significantly more compact than a fibered Michelson interferometer.L’objectif de cette thèse était de développer un capteur de surpression aérienne utilisant la rétro-injection optique dans une diode laser. Dans le cadre des expériences de détonique, les capteurs de surpression aérienne utilisés par le CEA Gramat sont en grande majorité piézoélectriques. Le but était de réaliser un capteur optique qui s’affranchit des limites intrinsèques des capteurs piézoélectriques notamment en termes de bande passante et de perturbations électromagnétiques. Dans le cadre d’une étude initiale menée au sein du laboratoire de recherches correspondant LICUR, entre le LAAS-CNRS et le CEA Gramat, la faisabilité d’un capteur par rétro-injection optique pour caractériser une onde de choc avait été évaluée. L’approche est basée sur la connaissance fine de l’effet acousto-optique aux très hautes pressions et des effets thermodynamiques induits par l’onde de choc. Le capteur génère des franges interférométriques, liées à la rétro-injection optique, proportionnelles aux variations d’indice le long du faisceau laser de mesure. Cette thèse décrit le dimensionnement complet du système capteur. Elle s’est déroulée en travaillant parallèlement sur le développement de traitement du signal et sur la validation progressive en conditions de laboratoire et réelles du capteur à rétro-injection optique. Les premiers essais ont démontré à la fois le fort potentiel de cette technologie et les difficultés inhérentes d’un capteur basé sur ce principe dont la détection des franges d’interférences à haute fréquence et la décorrélation de la mesure de pression à partir de l’indice de réfraction en prenant en compte les effets thermodynamiques. Ces essais ont été réalisés sur des moyens en laboratoire : enceinte miniature au LAAS et tube à choc dynamique au CEA Gramat. Ils ont permis de valider une première version du modèle acousto-optique (valable jusqu’à 20 bar) et de confronter la mesure de pression déduite avec les mesures temporelles obtenues via des capteurs piézoélectriques du commerce. Les expériences ont montré également que la forme du front de choc avait une influence sur la mesure optique. En effet, le capteur optique fait une intégration de l’indice optique le long du faisceau laser, il est donc nécessaire d’étudier l’influence du rayon de courbure afin d’obtenir la pression maximale du front de choc. Le traitement du signal produit par le capteur constitue également un point clé de cette thèse. La mise en place d’un algorithme, plus robuste que les précédents et basé sur un traitement par ondelettes, s’est avérée indispensable pour la détection des franges. Le capteur optique développé a été également déployé en conditions réelles à bas niveau sur un tube à choc d’une dizaine de centimètres de diamètre et sur des expériences de détonation champ libre à l’extérieur avec des niveaux de pression de l’ordre de 5 bar. Les derniers essais ont démontré la capacité du capteur optique à mesurer le front de montée de l’onde de choc, ainsi que la robustesse de sa conception opto-mécanique vis-à-vis des niveaux de pression, de la boule de feu, du flash lumineux et des températures extrêmes durant les expériences (-10°C ; +40°C). Enfin, une comparaison entre l’approche par rétro-injection optique et celle par l’interférométrie de Michelson fibrée a été menée, ces deux méthodes étant théoriquement capables de mesurer une variation de l’indice optique due à une onde de choc. Il a été montré que l’interféromètre de Michelson pouvait mesurer l’indice de réfraction avec une meilleure résolution (~10⁻⁸) mais était plus sensible aux réglages d’alignement de la visée laser que le capteur à rétro-injection optique, qui est cependant limité pour le moment à une frange d’interférence en termes de résolution (~10⁻⁵). Le capteur à rétro-injection optique est bien plus compact qu’un interféromètre de Michelson fibré
Comparative analysis of dynamic balance descriptors in humanoids and humans during perturbed bipedal locomotion and fall
No full textInternational audienceThis study identifies a robust parameter for quantifying instability in general biped systems by comparing three mechanical stability descriptors: the distance between the center of mass to the minimal moment axis (d CoM-MMA ), the margin of stability (MoS), and whole-body angular momentum (WBAM) in both humans and humanoid biped robots. We analyzed these metrics during normal and perturbed walking, including robot falls, a dynamic whose observation is limited in human trials due to safety concerns. Our comparative analyses demonstrate that d CoM-MMA is more predictive of different levels of instability and shows a clearer distinction between fall and non-fall states, compared to MoS and WBAM. These findings were consistent for both humans and biped robots, regardless of gait variability or the type and intensity of the perturbation methods. These qualities highlight its potential use in unified stability analysis in both fields, offering insights that can inform the design of exoskeletons, fall monitoring systems, and other gait-assistive devices for aging populations
Sufficient conditions for the absence of relaxation gaps in state-constrained optimal control
No full text9 page
Asymptotics of Ensemble Filters for Linear Stochastic Systems with Poisson-Sampled Observations
No full textFor continuous-time linear stochastic dynamical systems driven by Wiener processes, we consider the problem of designing ensemble filters when the observation process is randomly time-sampled. For the design of ensemble filters, we consider a class of continuous-discrete diffusion processes with additive Gaussian noise and several design parameters, which are used to describe the evolution of the individual particles in the ensemble. These particles are coupled through the empirical covariance, and in some cases empirical mean as well, and require less computations for implementation than the optimal ones based on solving Riccati differential equations. For different choices of parameters, we can recover some common design techniques from the literature. Our focus in this work is on analyzing the asymptotic (in time) performance of these filters for sufficiently large number of particles. Using appropriate analysis tools, we derive differential equations to describe the expectation of empirical mean and sample covariance of the ensemble filters with respect to the sampling process and noise. The solutions of these differential equations (describing empirical moments) are shown to converge asymptotically to the mean and covariance of the optimal filter under certain conditions on the mean sampling rate of the observation process, and as the number of particles tends to infinity
Decreased cytoplasmic crowding via inhibition of ribosome biogenesis can trigger Candida albicans filamentous growth
No full textInternational audienceThe human fungal pathogen Candida albicans undergoes a morphological transition from a budding yeast to a filamentous form, which is associated with pathogenesis. Various cues mediate this transition including intracellular reorganisation. The cytoplasm is densely packed with proteins including large macromolecular complexes, such as ribosomes, and hence molecular crowding can impact a range of cellular processes. However, the relationship between cytoplasmic molecular crowding and morphological growth states is unclear. Using a fluorescent microrheological probe and single particle tracking, we observed a striking decrease in molecular crowding during filamentous growth in C. albicans. Based on simulations, proteomics and structural data from in situ cryo-EM, we show that the reduction in crowding is due to a decrease in ribosome concentration that results in part from an inhibition of ribosome biogenesis, combined with an increase in cytoplasmic volume; leading to a dilution of ribosomes. Filamentation was enhanced in a mutant defective in ribosome biogenesis, whilst translation was not affected, suggesting that inhibition of ribosome biogenesis is a trigger for C. albicans morphogenesis. Overall we show that filamentous growth is associated with reduced cytoplasmic crowding via changes in ribosome concentration, suggesting that combination therapies in which ribosome biogenesis is also targeted may be advantageous
Planar Microwave Sensors: State of the Art and Applications
No full textInternational audienceThis review paper focuses on the latest advances and applications of planar microwave sensors contributed by the most prominent researchers in the field. The paper presents the different working principles, design approaches, fabrication technologies, materials, and applications of a wide variety of planar sensors operating at microwave and millimeter-wave frequencies, including contact and contactless sensors, wired and wireless sensors, microfluidic sensors, “green” sensors, wearables, biosensors, physical sensors, chemical sensors, and more. Advanced techniques for sensor performance optimization (e.g., sensitivity, resolution, selectivity, etc.), based on artificial intelligence, active feedback loops, microwave spectroscopy, losses engineering, etc., will also be discussed in the paper
Characterization Methods for Evaluating Charge Carrier Trapping Mechanisms of SiC Power MOSFETs
No full textInternational audienceThe SiC-oxide interface plays a key role in the dynamic and long-term reliability performance of SiC power MOSFETs. This paper outlines characterization approaches for evaluating charge carrier trapping mechanisms occurring within SiC Power MOSFETs that can increase understanding and validate common assumptions about gate-oxide-related reliability concerns of SiC power MOSFETs