Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Vinylcyclopropanes as All-Carbon 1,5-Dipoles: A Reactivity Switch for Palladium (5+4) Cycloadditions
No full textInternational audienc
Explainability in Deep Reinforcement Learning
No full textInternational audienceA large set of the explainable Artificial Intelligence (XAI) literature is emerging on feature relevance techniques to explain a deep neural network (DNN) output or explaining models that ingest image source data. However, assessing how XAI techniques can help understand models beyond classification tasks, e.g. for reinforcement learning (RL), has not been extensively studied. We review recent works in the direction to attain Explainable Reinforcement Learning (XRL), a relatively new subfield of Explainable Artificial Intelligence, intended to be used in general public applications, with diverse audiences, requiring ethical, responsible and trustable algorithms. In critical situations where it is essential to justify and explain the agent's behaviour, better explainability and interpretability of RL models could help gain scientific insight on the inner workings of what is still considered a black box. We evaluate mainly studies directly linking explainability to RL, and split these into two categories according to the way the explanations are generated: transparent algorithms and post-hoc explainaility. We also review the most prominent XAI works from the lenses of how they could potentially enlighten the further deployment of the latest advances in RL, in the demanding present and future of everyday problems
Set Propagation Techniques for Reachability Analysis
No full textInternational audienceReachability analysis consists in computing the set of states that are reachable by a dynamical system from all initial states and for all admissible inputs and parameters. It is a fundamental problem motivated by many applications in formal verification, controller synthesis, and estimation, to name only a few. This paper focuses on a class of methods for computing a guaranteed over-approximation of the reachable set of continuous and hybrid systems, relying predominantly on set propagation: starting from the set of initial states, these techniques iteratively propagate a sequence of sets according to the system dynamics. After a review on set representation and computation, the paper presents the state of the art on set propagation techniques for reachability analysis of linear, nonlinear, and hybrid systems. The paper ends with a discussion on successful applications of reachability analysis to real world-problems. Content
A Sensor-Based Data Analytics for Patient Monitoring in Connected Healthcare Applications
No full textInternational audienceNowadays, keeping a strong and good health is one of the main concern of the general public or governments. The Internet of Things (IoT) has been emerged as an efficient solution to build smart healthcare systems deployed either at hospitals or in-home. Such networks rely on biomedical sensors which are used in electronics-based medical equipment to remotely collect vital signs of patients (pressure, temperature, hart rate, oxygen saturation etc.). Generally, these biosensors are implemented on or inside the patient's body and take three types of record data such as numerical, images and videos. However, the big data collected by various biomedical sensors along with the need of emergency detection, the limited sensor energies, and the prediction of the progress of patient situation are the major challenges for heath-based IoT applications. In order to overcome these challenges, we propose, in this paper, an efficient sensor-based data analytics for real-time patient monitoring and assessment to help both hospital and medical staff. The proposed mechanism consists in three phases: Emergency detection, adapting sensing frequency and real time prediction of patient situation. Through simulations on real health data, we show the effectiveness of our mechanism compared to other exiting techniques
Imaging junctions of waveguides
No full textInternational audienceIn this paper we address the identification of defects by the Linear Sampling Method in half-waveguides which are related to each other by junctions. Firstly a waveguide which is characterized by an abrupt change of properties is considered, secondly the more difficult case of several half-waveguides related to each other by a junction of complex geometry. Our approach is illustrated by some two-dimensional numerical experiments
An extension of the proximal point algorithm beyond convexity
No full textInternational audienceAbstract We introduce and investigate a new generalized convexity notion for functions called prox-convexity. The proximity operator of such a function is single-valued and firmly nonexpansive. We provide examples of (strongly) quasiconvex, weakly convex, and DC (difference of convex) functions that are prox-convex, however none of these classes fully contains the one of prox-convex functions or is included into it. We show that the classical proximal point algorithm remains convergent when the convexity of the proper lower semicontinuous function to be minimized is relaxed to prox-convexity
A Modified RC-pLMS Adaptive Beamformer for Secure Digital Communication
No full textInternational audienceA modified reduced complexity parallel least mean square (mRC-pLMS) adaptive beamforming algorithm for high precision directivity and secure data communication is proposed in this paper. A high performance RC-pLMS algorithm has been proposed, recently, to eliminate the tradeoff between the LMS steady state error and its convergence speed while maintaining a low computational complexity structure. RC-pLMS is obtained by simplifying the two stages parallel LMS (pLMS) and adding a filter to the inputs, thus eliminating the need of an additional LMS filter. To further improve the RC-pLMS convergence speed and accuracy for fast and secure data communication we propose a modified RC-pLMS algorithm. mRC-pLMS is obtained by updating the RC-pLMS weight update equation to make use of the filtered input signal rather than the original input. Numerical simulations reflected by the mean square error convergence behavior and beam pattern, demonstrate the superior performance of the mRC-pLMS in providing faster convergence, lower steady state error and better interference attenuation while maintaining identical RC-pLMS resource requirements
Repetition rate performance for frequency mixing of four simultaneous QPSK signals based on a SOA-MZI photonic sampling mixer
No full textInternational audienceA simultaneous frequency up conversion of four intermediate frequency (IF) signals is carried out by utilizing a semiconductor optical amplifier Mach-Zehnder interferometer (SOA-MZI) in a differential configuration for radio over fiber applications. A sampling signal compelled by an optical pulse clock source produces 10-ps-width pulses at a repetition rate domain that is from 7.8 to 19.5 GHz. The four IF signals carrying quadratic phase shift keying (QPSK) data at frequencies fm are up converted at the SOA-MZI output at mixing frequencies nfsk ± fm, where k and m equal 1, 2, 3, and 4 and n is the harmonic rank of the sampling signal. The simulation study for simultaneous frequency up conversion relied on the SOA-MZI sampling mixer is developed to acquire the conversion gain and the error vector magnitude (EVM) in the repetition rate range. Using the virtual photonics integrated simulator, we show that incrementing the repetition rate from 7.8 to 19.5 GHz improves the competence and merit of the optical transmission system due to a better signal level and a lower aliased noise power with a higher sampling rate. Positive conversion gains were achieved at a higher mixing frequency for each channel. Concomitantly, the benefit on the conversion gain provided by augmenting the sampling frequency is 14 dB. By increasing the repetition rate, the EVM can be ameliorated up to 12% for all channels. In addition, it degrades more when the frequency channel increases over the repetition rate range. The maximum bit rate of 25 Gbit/s with a QPSK modulation meets the forward error correction limi
