Archive ouverte HAL-LAAS
Not a member yet
12189 research outputs found
Sort by
Sur l'impact des distorsions dans l'imagerie hyperspectrale codée
No full textInternational audienceSingle Disperser Coded Aperture Spectral Snapshot Imagers (SD-CASSI) have been widely used to reduce the data quantity of hyperspectral images (HSI) through coded acquisitions. Based on prior work [1], most SD-CASSI systems include an expensive double-Amici prism as their dispersive element, as it allows for a direct-view and almost no distortions, contrary to systems relying on off-the-shelf single prism. We compared those two types of systems on a reconstruction task and concluded that distortions can be compensated as long as they are known and modeled.Les Single Disperser Coded Aperture Spectral Snapshot Imagers (SD-CASSI) sont très utilisés pour réduire la quantité de données des images hyperspectrales grâce à des acquisitions codées. Sur la base de travaux antérieurs [1], la plupart des systèmes SD-CASSI comprennent un prisme Amici coûteux comme élément dispersif, car il permet d'avoir un système aligné et pratiquement aucune distorsion, contrairement aux systèmes qui utilisent un prisme simple, disponible dans le commerce. Nous avons comparé ces deux types de systèmes pour une tâche de reconstruction et avons conclu que les distorsions peuvent être compensées tant qu'elles sont connues et modélisées
Efficient First-Principles EPR g-Tensor Calculations with QE-CONVERSE: Enabling Large Supercells via the Single-Point Formula
No full textNational audienc
Scheduling Data Transfers in Space Missions with Priorities and Interruptions: Technical Report
No full textIn deep space missions, scientific data generated by onboard instruments must be temporarily stored in local memory buffers before being downlinked to Earth during limited communication windows. Efficient scheduling of these data transfers is essential to prevent buffer overflows and data loss, particularly in the presence of uncertainties. Previous work has considered the overlapping Memory Dumping Problem (oMDP), which consists in assigning transfer priorities to the memory buffers and minimize the peaks memory usage, which reduces the risk of overflow. In this paper, we consider additional decisions in the transfer plans that are implementable in practice: data transfer from each buffer can be interrupted after a given time, once per downlink window, preventing it from dumping data until the next window, but redistributing the unused bandwidth to the other buffers. The new problem is called oMDPi (oMDP with interruptions). We obtain new complexity results, showing that oMDPi is NP-complete for at least two windows. While the complexity status of the single window oMDPi remains open, we propose a polynomial-time heuristic to solve it. We propose a hybrid heuristic to solve the general oMDPi, embedding a flow relaxation and a single window heuristic. The results on both real and realistic generated instances show that our heuristic achieves a significant reduction of memory peaks in a reasonable time compared to previous works, making the new policy attractive for future space missions
Apprentissage par renforcement avec exploration rapide pour les applications aux réseaux stochastiques
No full textInternational audienceWe introduce FVRL (Fleming–Viot reinforcement learning), a reinforcement learning algorithm for optimisation problems where a long-term objective is largely influenced by states that are very rarely observed under all policies. In this context, usual discovery techniques including importance sampling are inapplicable because no alternative policy exists that increases the observed frequency of the rare states. We instead propose a novel approach that uses Fleming–Viot particle systems, a family of stochastic processes evolving simultaneously under the same law, that exploits prior knowledge of the environment to boost exploration of the rare states. A renewal theory argument allows us to consistently estimate the stationary probability of the rare states from excursions that have considerably lower sample complexity than usual Monte Carlo explorations. We demonstrate how to combine this estimator with policy gradient learning to construct the FVRL algorithm, which is suited to efficiently solve problems where the optimisation function is expressed as a long-run expectation, such as the long-run expected reward. We show that the FVRL algorithm converges to a local optimiser of the parameterised objective function, and illustrate the method on two optimisation problems that aim at minimising the long-run expected cost under admission control policies of threshold type: a simple M/M/1 queue system and a two-job-class loss network. Our experimental results show that, under the same sample complexity, FVRL outperforms a vanilla Monte Carlo reinforcement learning method by converging to the optimum thresholds considerably faster
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
How Machine Learning can contribute to evaluate stress level
No full textInternational audiencesolution has been embedded or run in real-time. Finally, none of them uses a database combination that can be compared with the literature (public database) while being robust for real-life use (customized database).The Pulse system, introduced in [11], addresses embedded and real-time challenges by using Ballistocardiography [12] and includes a strain sensors matrix and dedicated circuits. A real-time embedded software called BCG-VARS [13] associated with Pulse performs contactless monitoring of vital parameters across various equipment types. Unlike ECG or PPG, this solution does not need to be worn, providing greater comfort and ease of use. The software features two algorithms and is designed to extract three vital parameters: the actigraphy which is a quantification of movement, Breath-to-Breath Interval (BBI) which is the distance between 2 breath cycles, and heart Inter-Beat Interval (IBI) which is the distance between 2 heart cycles. BCG-VARS and Pulse are low-cost and can be embedded.This paper introduces a novel approach for contactless stress detection using machine learning techniques. We propose an innovative algorithm, VESLA (Variable and Embedded Stress Level Algorithm), designed to monitor stress levels dynamically. While existing databases typically employ binary stress labels, VESLA enables the quantification of multiple stress levels, offering a flexible solution adaptable to various applications and surpassing many current models in the literature. Due to the unobtrusive design of the Pulse system, longterm stress evaluation becomes feasible, allowing for chronic stress monitoring without inducing a "white coat" effect. This solution can be effectively integrated into equipment in airplanes, emergency services, and teleconsultation settings to provide continuous, reliable stress monitoring.This article is organized as follows: Section II details the database generation required for ML training. Section III describes the training and evaluation of ML models for absolute stress measurement. Section IV highlights performance improvements achieved through relative stress measurement with the VESLA algorithm, which enhances the ML model. Finally, the paper concludes with a discussion of the models' performance and overall findings. II. DATABASE GENERATION BASED ON A PUBLIC DATASETCreating datasets for ML models is often the most timeconsuming stage, especially when involving human subjects.Abstract-Chronic stress, common in modern lifestyles, contributes to heart disease, hypertension, diabetes, and obesity, highlighting the need for effective stress management. In this article, stress detection using machine learning (ML) techniques has been explored. However, individuals have very different stress levels. Therefore, measuring the evolution of stress over time is more relevant. So, an innovative algorithm called VESLA (Variable and Embedded Stress Level Algorithm) has been developed to assess stress evolution. Despite binary nature labels in the dedicated database, VESLA can quantify multiple stress levels, adapting to the application's requirements. This algorithm outperforms literature models, achieving an average accuracy of 94.1 % with an open-access database
3604 Artificial Intelligence-driven failure prediction on TomoTherapy® systems
No full textInternational audienc
Scheduling Data Transfer with Priorities and Interruptions for Space Missions
No full textInternational audienceIn the context of space missions, an efficient data transfer strategy is crucial to avoid onboard buffer overflow. In many missions, such as the ESA's Rosetta mission, the data produced by each instrument is temporarily stored in a dedicated buffer and then dumped to Earth during downlink windows under a limited bandwidth. Previous work has considered the overlapping Memory Dumping Problem (oMDP), which consists in assigning transfer priorities to the memory buffers and minimize the peaks memory usage. The oMDP is NP-hard and has been tackled using heuristic methods that demonstrated their efficiency on real instances. In this paper, we consider additional decisions in the memory dumping plans that are implementable in practice: data transfer from each buffer can be interrupted after a given time, once per downlink window, preventing it from dumping data until the next window. The new problem is called oMDPi (oMDP with interruptions). We obtain several structural properties and complexity results, showing in particular that oMDPi is NP-hard for at least two windows. While the complexity status of the single window oMDPi remains open, we propose a polynomial-time heuristic to solve it. We propose a hybrid heuristic to solve the general oMDPi, embedding a flow relaxation and the single window heuristic. The results on both real and realistic generated instances show that our heuristic achieves a significant reduction of memory peaks in a reasonable time compared to previous works, making the new policy attractive for future space missions
Optimal Motion Prediction for Human-to-Robot Handovers
No full textInternational audienceSeamless human-robot handovers require precision, timing, and safety. In the absence of visual feedback for humans, robots rely on accurately estimating and predicting their motion. In this work, a real-time human motion prediction and estimation framework for human-to-robot handovers relying on a planar biomechanical model and cost functions extracted from the motor control literature is proposed. Thanks to inverse reinforcement learning, it is possible to iteratively determine the optimal weighting of these cost functions by solving a direct optimal control problem for reaching tasks. An affordable, markerless human pose estimation pipeline was used to estimate in real-time and predict the human arm motion. These predictions were then integrated into a model predictive controller for a seven-degree-of-freedom robot manipulator, successfully intercepting participants' hands in 88.6 ± 8.0% of trials, 0.63s before they reached their intended final hand pose. Experimental validation with blindfolded participants resulted in a predicted joint angle error of 8.7±4.6deg during handover trials. The proposed framework offers a promising solution for safe and effective human-to-robot handovers, particularly for applications involving visually impaired users
