Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Constrained Hierarchical Clustering via Graph Coarsening and Optimal Cuts
No full text5 pages, appeared at the Asilomar Conference on Signals, Systems, and Computer, 11/2023International audienceMotivated by extracting and summarizing relevant information in short sentence settings, such as satisfaction questionnaires, hotel reviews, and X/Twitter, we study the problem of clustering words in a hierarchical fashion. In particular, we focus on the problem of clustering with horizontal and vertical structural constraints. Horizontal constraints are typically cannot-link and must-link among words, while vertical constraints are precedence constraints among cluster levels. We overcome state-of-the-art bottlenecks by formulating the problem in two steps: first, as a soft-constrained regularized least-squares which guides the result of a sequential graph coarsening algorithm towards the horizontal feasible set. Then, flat clusters are extracted from the resulting hierarchical tree by computing optimal cut heights based on the available constraints. We show that the resulting approach compares very well with respect to existing algorithms and is computationally light
Investigation on the dual use of thermal insulation materials as blast mitigation solution
No full textInternational audienc
Progress in Hybrid Plasma Wakefield Acceleration
No full textInternational audiencePlasma wakefield accelerators can be driven either by intense laser pulses (LWFA) or by intense particle beams (PWFA). A third approach that combines the complementary advantages of both types of plasma wakefield accelerator has been established with increasing success over the last decade and is called hybrid LWFA→PWFA. Essentially, a compact LWFA is exploited to produce an energetic, high-current electron beam as a driver for a subsequent PWFA stage, which, in turn, is exploited for phase-constant, inherently laser-synchronized, quasi-static acceleration over extended acceleration lengths. The sum is greater than its parts: the approach not only provides a compact, cost-effective alternative to linac-driven PWFA for exploitation of PWFA and its advantages for acceleration and high-brightness beam generation, but extends the parameter range accessible for PWFA and, through the added benefit of co-location of inherently synchronized laser pulses, enables high-precision pump/probing, injection, seeding and unique experimental constellations, e.g., for beam coordination and collision experiments. We report on the accelerating progress of the approach achieved in a series of collaborative experiments and discuss future prospects and potential impact
Turbo Equalization for Underwater Communication Systems using Rotated Constellations
No full textInternational audienceThe available bandwidth of the underwater acoustic channel is severely limited. Therefore, it is of interest to use modulations with high spectral efficiency, such as rotated constellations. These constellations offer better theoretical performance than conventional constellations, without reducing the spectral or energy efficiency of the system. However, in practice, when conventional equalization algorithms are employed for a system using rotated constellations, the theoretical gains are not fully achieved. To address this issue, this paper proposes a new turbo equalization algorithm that inherently considers the characteristics of rotated signals and allows for a significant gain compared to conventional methods, as demonstrated by simulations, in particular over a real underwater channel
Time Blocks Decomposition of Multistage Stochastic Optimization Problems
No full textInternational audienceMultistage stochastic optimization problems are, by essence, complex as their solutions are indexed both by stages and by uncertainties. Their large scale nature makes decomposition methods appealing, like dynamic programming which is a sequential decomposition using a state variable defined at all stages. In this paper, we introduce the notion of state reduction by time blocks, that is, at stages that are not necessarily all the original stages. Then, we prove a reduced dynamic programming equation. We position our result with respect to the most well-known mathematical frameworks for dynamic programming. We illustrate our contribution by showing its potential for applied problems with two time scales
Optimisation des grands systèmes: Méthodes de décomposition/coordination dans le cadre déterministe
No full textMasterLe cours "Optimisation des grands systèmes" a été donné durant de nombreuses années à l'ENSTA comme enseignement de troisième année dans le parcours dédié à l'approfondissement en optimisation et recherche opérationnelle. Ces notes constituent le support de ce cours et ont pour but de fournir aux étudiants l'essentiel de ce qu'il faut retenir du cours. Le livre "Décomposition-coordination en optimisation déterministe et stochastique", publié en 2017 chez Springer, reprend et détaille l'ensemble des notions du cours, et présente l'extension de ces notions au cas stochastique.L'objectif de ce cours est de présenter les méthodes mathématiques et algorithmes permettant d'optimiser un système dont la taille et l'hétérogénéité sont telles que les méthodes "classiques" de l'optimisation ne peuvent pas être mises en oeuvre.Dans ce cours, on se limite à la présentation des méthodes de décomposition et coordination dans le cadre de l'optimisation convexe différentiable déterministe.Le cours comprend essentiellement deux parties. Dans la première partie, on cherche à introduire les idées de la décomposition/coordination et à développer les interprétations économiques sur un modèle simple, sans se préoccuper outre mesure de généralité ou de rigueur mathématique. Dans la deuxième partie, on développe une théorie générale basée sur le principe du problème auxiliaire, permettant d'une part de lever les restrictions qui paraissaient essentielles dans la première partie, et d'autre part d'étudier dans un cadre unifié la convergence des algorithmes de coordination. Enfin, un exemple correspondant à un réseau de distribution d'eau potable de grande taille est présenté et sert à illustrer l'ensemble des méthodes présentés durant le cours
Asymmetric synthesis of a stereopentade fragment toward latrunculins
No full textInternational audienceLatrunculins are marine toxins used in cell biology to block actin polymerization. The development of new synthetic strategies and methods for their synthesis is thus important in order to improve, modulate or control this biological value. The total syntheses found in the literature all target similar disconnections, especially an aldol strategy involving a recurrent 4-acetyl-1,3-thiazolidin-2-one ketone partner. Herein, we describe an alternative disconnection and subsequent stereoselective transformations to construct a stereopentade amenable to latrunculin and analogue synthesis, starting from (+)-β-citronellene. Key stereoselective transformations involve an asymmetric Krische allylation, an aldol reaction under 1,5- anti stereocontrol, and a Tishchenko–Evans reduction accompanied by a peculiar ester transposition, allowing to install key stereogenic centers of the natural products
Semi-analytical load models describing the progressive immersion of a fixed vertical cylinder in a breaking wave
No full textInternational audienceThis paper is part of the DIMPACT (Design of floating wind turbines and impacts of energetic steep and breaking waves) project concerning slamming loads on floating offshore wind turbines. Two semi-analytical models accounting for the progressive immersion of a vertical cylinder in a breaking wave are presented. These models are based on the rate of change of the fluid momentum and the continuity of the added mass during immersion. The first model is based on the Generalized von Kármán Model where the real shape of the body and its added mass are accounted for. To account for the nonlinearity of the flow kinematics, the second-order terms of the momentum equation are considered in the inertia load. The total load is obtained by adding the drag force using a variable drag coefficient. To better predict the slamming load, a second model based on Wagner's theory is presented. The Modified Logvinovich Model provides the pressure distribution at the first instants of impact. It is assumed that the flow does not separate from the structure. The load models are applied in a strip-theory approach under the Froude–Krylov assumption. For comparison, a Navier–Stokes solver is used to generate four phase-focused breaking waves. The resulting free surface shape and the ambient kinematics of the two-dimensional waves are used in the semi-analytical models. The present models are compared with existing slamming load formulations and a three-dimensional numerical simulation with an actual cylinder, where the load on four sections of the cylinder are considered. The force in the deeply immersed sections is in good agreement in all models considered. In the partially submerged section, the proposed load formulations and the three-dimensional simulation predict a similar load. Other standard engineering formulas predict a much lower force in this section. In the area impacted by the crest, the Froude–Krylov-based approaches predict a much higher force than that given by the Navier–Stokes result, likely due to the free surface being disturbed by the cylinder
Incentives and co-evolution: Steering linear dynamical systems with noncooperative agents
No full textInternational audienceModern socio-technical systems, such as smart energy grids, ride-hailing services, or digital marketplaces, typically consist of many interconnected users and competing service providers. Within these systems, notions like market equilibrium are tightly connected to the ``evolution'' of the network of users. In this paper, we model the users' state and dynamics as a linear dynamical system, and the service providers as agents taking part to a generalized Nash game, whose outcome coincides with the input of the users' dynamics. We are thus able to characterize the notion of co-evolution of the market and the network dynamics and derive conditions leading to a pertinent notion of equilibrium. These conditions are based on dissipativity arguments and yield easy-to-check linear matrix inequalities. We then turn the problem into a control one: how can we incentivize or penalize the service providers acting as little as possible to steer the whole network to a desirable outcome? This so-called light-touch policy design problem can be solved through bilinear matrix inequalities. We also provide a dimensionality-reduction procedure, which offers network-size independent conditions and design tools. Finally, we illustrate our novel notions and algorithms on a simulation setup stemming from digital market regulations for influencers, a topic of growing interest