Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Approximations variationnelles bayésiennes pour le traitement du signal: quelques applications en parcimonie et reconstruction de phase
No full textAs an alternative to sampling methods, Bayesian variational approximations are a family of approximations that approximate probability distributions by others, with desirable factorizations. Implemented in low complexity algorithmic procedures, they allow to obtain an analytical expression (although approximated) of the target distributions. These methods are successfully considered for solving inverse problems in Bayesian frameworks.At the heart of my research, Bayesian variational approximations have notably proven their value in two main types of problems :— (structured) sparse decompositions,— (informed) phase retrieval.My research has allowed me to approach these two major problems from two points of view : a methodological point of view, focusing on the development of new solutions based on Bayesian variational approximations, and an applied point of view, implementing these solutions in concrete practical settings. Among these applications, underwater acoustics holds an important place, notably through antenna processing in fluctuating oceanic environments or modal estimation in ultra-low frequencies and shallow-water environments.This work opens the way to numerous research perspectives, both methodological and applicative, combining Bayesian modeling, machine learning and taking into account the physics of the phenomena considered.Alternatives aux méthodes d’échantillonnage, les approximations variationnelles bayésiennes constituent une famille d’approximations approchant des distributions de probabilité par d’autres, présentant des factorisations désirables. Mises en oeuvre dans des procédures algorithmiques peu complexes, elles permettent d’obtenir une expression analytique (quoique approchée) des distributions cibles. Ces méthodes sont considérées avec succès pour la résolution de problèmes inverses dans des cadres bayésiens.Fil conducteur de mes recherches, les approximations variationnelles bayésiennes ont notamment prouve leur intérêt dans deux grands types de problèmes :— les décompositions parcimonieuses (structurées),— la reconstruction de phase (informée).Mes recherches m’ont permis d’aborder ces deux grands problèmes selon deux points de vue : un point de vue méthodologique, s’intéressant au développement de nouvelles solutions basées sur des approximations variationnelles bayésiennes, et un point de vue appliqué, mettant en œuvre ces solutions dans des cadres pratiques concrets. Parmi ces applications, l’acoustique sous-marine tient une place importante, `a travers notamment le traitement d’antenne en milieu océanique fluctuant ou l’estimation modale en ultra-basses fréquences et milieux petit-fond.Ces travaux ouvrent la voie `a de nombreuses perspectives de recherche tant méthodologiques qu’applicatives, combinant modélisation bayésienne, apprentissage automatique et prise en compte de la physique des phénomènes considérés
Classification de cornées humaines saines et pathologiques par la quantification d’images SD-OCT cliniques
No full textInternational audienc
Poly(Lactic Acid) (PLA)-Based Nanocomposites: Impact of Vermiculite, Silver, and Graphene Oxide on Thermal Stability, Isothermal Crystallization, and Local Mechanical Behavior
No full textInternational audienceThe structural, thermal, and mechanical properties of unreinforced and reinforced polylactic acid (PLA) were investigated. The PLA was a biopolymer that was reinforced with four fillers (i.e., graphene oxide (GO) and silver (Ag); vermiculite (VMT) and silver (Ag); and two organically modified vermiculites). The processing technique for the production of the composite materials were carefully planned. The PLA nanocomposites were investigated by examining their morphological aspects, changes in PLA phases and transitions and, most importantly, the effect on certain final properties. X-ray diffraction and differential scanning calorimetry (DSC) analysis indicated that the sample was completely amorphous. Thermogravimetric analysis (TGA) results indicated that the presence of reinforcing particles in the PLA matrix did not affect the thermal degradation of these composites. Furthermore, the local mechanical properties were investigated using the microindentation method to evaluate the effect of different nanofillers. Scanning electron microscopy (SEM) and a VHX-500 optical digital microscope (Keyence International, Mechelen, Belgium) were also used to examine the surface morphology of the PLA polymer composites. These results can help to select suitable fillers to enhance the PLA performance of biopolymers
Simultaneous Control and Guidance of an AUV Based on Soft Actor–Critic
No full textInternational audienceThe marine environment is a hostile setting for robotics. It is strongly unstructured, uncertain, and includes many external disturbances that cannot be easily predicted or modeled. In this work, we attempt to control an autonomous underwater vehicle (AUV) to perform a waypoint tracking task, using a machine learning-based controller. There has been great progress in machine learning (in many different domains) in recent years; in the subfield of deep reinforcement learning, several algorithms suitable for the continuous control of dynamical systems have been designed. We implemented the soft actor–critic (SAC) algorithm, an entropy-regularized deep reinforcement learning algorithm that allows fulfilling a learning task and encourages the exploration of the environment simultaneously. We compared a SAC-based controller with a proportional integral derivative (PID) controller on a waypoint tracking task using specific performance metrics. All tests were simulated via the UUV simulator. We applied these two controllers to the RexROV 2, a six degrees of freedom cube-shaped remotely operated underwater Vehicle (ROV) converted in an AUV. We propose several interesting contributions as a result of these tests, such as making the SAC control and guiding the AUV simultaneously, outperforming the PID controller in terms of energy saving, and reducing the amount of information needed by the SAC algorithm inputs. Moreover, our implementation of this controller allows facilitating the transfer towards real-world robots. The code corresponding to this work is available on GitHub
Why do you think this joke told by robot is funny? The humor style matters
No full textInternational audienc
Level-crossing distributions of kinematic variables in multidirectional second-order ocean waves
No full textInternational audienceThe conditional value of a stationary random process, given the level-upcrossing of another dependent stationary random process, is considered. Assuming that both processes are weakly non-Gaussian, an analytical approximation for the related conditional distribution is derived. It is based on a trivariate Edgeworth expansion truncated to non-Gaussian terms of lowest order, to which Rice's formula is then applied. As an application, the effect of level-upcrossing conditioning in second-order ocean waves is investigated. Upcrossing events are monitored for the sea surface elevation. The conditional distributions of different kinematic variables, given upcrossing, are considered for different sea-state configurations. Predictions from the analytical model are compared with numerical data obtained from Monte Carlo experiments. It is found that the analytical approximation provides conditional mean and variance in good agreement with numerical data, although moderate discrepancies appear for the sea states with the most severe wave steepnesses. Regarding the conditional skewness, given upcrossing, results are mixed, with significant discrepancies between the analytical approximation and numerical estimates, in a number of cases. An Edgeworth-type approximation is also provided for the upcrossing frequency and compared with Monte Carlo estimates; this analytical estimate is found to be accurate over a wide range of crossing levels
Element-selective analysis of ultrafast demagnetization in Co/Pt multilayers exhibiting large perpendicular magnetic anisotropy
No full textInternational audienceWe use time-resolved resonant magnetic scattering experiments to investigate the laser induced demagnetization of Co/Pt multilayers with large perpendicular magnetic anisotropy. Our pump-probe approach, which relies on extreme ultraviolet pulses obtained via high-harmonic generation, allows us to monitor the temporal evolution of the system in an element-selective fashion, thus to disentangle the contributions from permanent and proximity-induced magnetic moments. By analyzing the data gathered at the Co and Pt O3 edges, we are able to show that after femtosecond laser irradiation, both components display a very similar loss of magnetic order, with almost identical demagnetization amplitudes and characteristic times, but a small delay (10–20 fs) of the Pt magnetization quenching with respect to Co. Comparing our experiments with literature results, therefore, suggests that in these 3d–5d multilayers and alloys, the different constituents remain strongly coupled after optical excitation; no matter whether the thin films possess a preferred in-plane or out-of-plane easy magnetization direction
On Monocular Depth Estimation and Uncertainty Quantification using Classification Approaches for Regression
No full textInternational audienceMonocular depth is important in many tasks, such as 3D reconstruction and autonomous driving. Deep learning based models achieve state-of-the-art performance in this field. A set of novel approaches for estimating monocular depth consists of transforming the regression task into a classification one. However, there is a lack of detailed descriptions and comparisons for Classification Approaches for Regression (CAR) in the community and no in-depth exploration of their potential for uncertainty estimation. To this end, this paper will introduce a taxonomy and summary of CAR approaches, a new uncertainty estimation solution for CAR, and a set of experiments on depth accuracy and uncertainty quantification for CAR-based models on KITTI dataset. The experiments reflect the differences in the portability of various CAR methods on two backbones. Meanwhile, the newly proposed method for uncertainty estimation can outperform the ensembling method with only one forward propagation
Wireless Communication Attack Using SDR and Low-Cost Devices
No full textInternational audienceWhile Wireless communication (WLC) enhances the user mobility and extends the connection and services to extreme isolated points, is the only way to establish a connection over long distance (mainly earth and space), it has many weak points: Interference, Spectrum limitations, bandwidth cost, various regulations, etc. But the major drawback is the security and risk vulnerabilities. To clarify this idea and highlight several security issues, we are working on the weak points of our wireless networks and communication protocols. In this manuscript, we develop several scenarios of Wireless attacks using simple and low-price equipment. Indeed using Software Defined Radio (SDR), a potential hacker can now access a wide range of wireless-based communication like Keyless entry, GPS and RFID system. It can also interfer and jam several other WLC services and networks