Archive ouverte HAL-LAAS
Not a member yet
12189 research outputs found
Sort by
De la lumière au noir, à travers la BD et l'œuvre de Soulages - Agrocampus RODEZ
No full textIntervention devant une classe de seconde menant un projet sur l'œuvre de Pierre Soulages. Lycée de La Roque - Agrocampus de RODE
Approximate D-optimal design and equilibrium measure
No full textWe introduce a minor variant of the approximate D-optimal design of experiments with a more general information matrix that takes into account the representation of the design space S. The main motivation (and result) is that if S in R^d is the unit ball, the unit box or the canonical simplex, then remarkably, for every dimension d and every degree n, one obtains an optimal solution in closed form, namely the equilibrium measure of S (in pluripotential theory). Equivalently, for each degree n, the unique optimal solution is the vector of moments (up to degree 2n) of the equilibrium measure of S. Hence finding an optimal design reduces to finding a cubature for the equilibrium measure, with atoms in S, positive weights, and exact up to degree 2n. In addition, any resulting sequence of atomic D-optimal measures converges to the equilibrium measure of S for the weak-star topology, as n increases. Links with Fekete sets of points are also discussed. More general compact basic semi-algebraic sets are also considered, and a previously developed two-step design algorithm is easily adapted to this new variant of D-optimal design problem
Integer and Constraint Programming for the Offline Nanosatellite Partition Scheduling Problem
No full textInternational audienceEffective scheduling of tasks for nanosatellites is essential, given their limited onboard resources and capabilities. In a typical nanosatellite mission, the onboard computer has to run payload tasks linked to the mission objective, such as observations and measurements for science campaigns, but also communication and avionic tasks needed for navigation and safety. We consider a partitioned real-time system where a partition is a job that embeds a set of elementary tasks implementing one of the above-described functions. The offline nanosatellite partition scheduling problem consists in scheduling a set of partitions on a single core processor within a fixed time frame that will be repeated cyclically, while maximizing the number and duration of scheduled payload partitions. The paper first establishes similarities and differences with related scheduling problems. We then prove that the problem is strongly NP-hard. A Mixed Integer Linear Programming (MILP) matheuristic and a Constraint Programming (CP) model are proposed. We compare the MILP and CP approaches in a multi-objective context and demonstrate the relevance of the latter to solve the problem efficiently both on randomly generated instances and on a real nanosatellite case study of the University Space Center of Toulouse: the NIMPH project. The CP model is embedded in NanoSatScheduler, an open source software with a user interface designed for the offline nanosatellite partition scheduling problem. For the NIMPH real-case study, the proposed solution outperforms the semi-manual approach used so far. As a result, the NIMPH team has adopted NanoSatScheduler as an operational tool for their mission
Review on cavity-resonant integrated grating filters
No full textInternational audienceWe review the investigations that have been carried out over the past decade on waveguide Fabry-Pérot microcavities with an embedded input/output grating coupler, devices often referred as Cavity-Resonant Integrated Grating Filters (CRIGFs). We will explain how the device geometry and use of various technological platforms (SiON, LiNbO3 on insulator, GaAs) has permitted the fabrication of spatially-localized wavelength filters that can operate from the nearinfared to mid-infrared, subsequently enabling applications as laser spectral stabilization or as pixelated filter for (hyperspectral) imaging. Furthermore, we will show that the design can be adjusted to obtain critically-coupled highquality factor microresonators with a view to induce efficient second harmonic generation or, more generally, nonlinear parametric conversion. Finally, we will present our most recent investigations that have been devoted to the selective excitation of the supported higher order spatial modes and how the latter can be used to implement reconfigurable logical gates
Simulation of a GaInP 2 -GaAs -GaInAsN -Ge photovoltaic cell for space applications
No full textInternational audienceWe use an open-source simulation software to design III-V multi-junction solar cells containing a dilute nitride subcell. We discuss the different modeling techniques in terms of physical models and numerical methods, and provide an educated guess for a GaInP2 -GaAs -GaInAsN -Ge lattice-matched 4-junction solar cell performance assessment. We suggest a way to ensure an effective computation of the main output values of an opensource Poisson-Drift-Diffusion solver for multi-junction solar cell analysis. An optimization of a multi-junction cell including a n-i-p subcell is performed, starting from basic theoretical considerations, to the use of a double-layer anti-reflective coating to target the most limiting subcell, and then with an optimization of the n-i-p architecture in the 1eV subcell with respect to the GaInAsN layer properties. We show that the detrimental effects of low diffusion lengths in the GaInAsN layer can be alleviated by taking advantage of field-assisted collection
FEMDA: Un framework unifié pour l'analyse discriminante
No full textInternational audienceAlthough linear and quadratic discriminant analysis are widely recognized classical methods, they can encounter significant challenges when dealing with non-Gaussian distributions or contaminated datasets. This is primarily due to their reliance on the Gaussian assumption, which lacks robustness. We first explain and review the classical methods to address this limitation and then present a novel approach that overcomes these issues. In this new approach, the model considered is an arbitrary Elliptically Symmetrical (ES) distribution per cluster with its own arbitrary scale parameter. This flexible model allows for potentially diverse and independent samples that may not follow identical distributions. By deriving a new decision rule, we demonstrate that maximum-likelihood parameter estimation and classification are simple, efficient, and robust compared to state-of-the-art methods
Self-assembled GaAs quantum dashes for direct alignment of liquid crystals on a III-V semiconductor surface
No full textInternational audienceThe development of tunable photonic devices is strategic for miniaturized optical instrumentation and sensing systems. Exploiting the birefringence variation of liquid crystals (LCs) instead of MEMS actuation in such devices could bring better spectral stability and lower power consumption. However, aligning LCs inside a III-V semiconductor device is tricky. We demonstrate that self-assembled gallium arsenide (GaAs) quantum dashes (QDHs) could serve as direct planar aligners for LC nematic molecules. The alignment quality and birefringence variation of a LC-microcell embedding QDHs are shown to be similar to those of a polymer nanograting-based reference, with the added advantage of better electrical performance
Wheelchair caster power losses due to rolling resistance on sports surfaces
No full textInternational audienceThe gross mechanical efficiency of the manual wheelchair propulsion movement is particularly low compared to other movements. The energy losses in the manual wheelchair propulsion movement are partly due to energy losses associated with the wheelchair, and especially to the rolling resistance of the wheels. The distribution of mass between the front rear wheels and the caster wheels has a significant impact on the rolling resistance. The study of the caster wheels cannot therefore be neglected due to their involvement in rolling resistance. Thus, this study aimed to evaluate the power dissipated due to rolling resistance by different caster wheels, at different speeds and under different loadings on various terrains. Four caster wheels of different shapes, diameters, and materials were tested on two surfaces representative of indoor sports surfaces at four different speeds and under four loadings. The results showed a minimal dissipated power of 0.4 ± 0.2 W for the skate caster, on the parquet, at 0.5 m/s and under a loading of 50 N. The maximal mean power dissipated was 43.3±27.6 W still for the skate caster, but on the Taraflex, at 1.5 m/s and under loading of 200 N. The power dissipated on the parquet was lower than the one on the Taraflex. The Spherical and Omniwheel caster wheels dissipated less power than the two other casters. This study showed that caster wheels cannot be neglected in the assessment of gross mechanical efficiency, particularly in light of the power dissipated by athletes during propulsion.</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
Towards Zero-Shot Cross-Agent Transfer Learning via Latent-Space Universal Notice Network
No full textInternational audienceDespite numerous improvements regarding the sample-efficiency of Reinforcement Learning (RL) methods, learning from scratch still requires millions (even dozens of millions) of interactions with the environment to converge to a high-reward policy. This is usually because the agent has no prior information about the task and its own physical embodiment. One way to address and mitigate this data-hungriness is to use Transfer Learning (TL). In this paper, we explore TL in the context of RL with the specific purpose of transferring policies from one agent to another, even in the presence of morphology discrepancies or different stateaction spaces. We propose a process to leverage past knowledge from one agent (source) to speed up or even bypass the learning phase for a different agent (target) tackling the same task. Our proposed method first leverages Variational Auto-Encoders (VAE) to learn an agent-agnostic latent space from paired, time-aligned trajectories collected on a set of agents. Then, we train a policy embedded inside the created agent-invariant latent space to solve a given task, yielding a task-module reusable by any of the agents sharing this common feature space. Through several robotic tasks and heterogeneous hardware platforms, both in simulation and on physical robots, we show the benefits of our approach in terms of improved sample-efficiency. More specifically we report zero-shot generalization in some instances, where performances after transfer are recovered instantly. In worst case scenarios, performances are retrieved after fine-tuning on the target robot for a fraction of the training cost required to train a policy with similar performances from scratch