HAL - Lille 3
Not a member yet
5967 research outputs found
Sort by
Non-quadratic stabilization of switched affine systems
No full textInternational audienceIn this paper we consider the problem of non-quadratic stabilization of switched affine systems. Using a switching Lyaponuv function, LMI conditions based constructive method are proposed in order to design nonlinear switching surfaces and provide an estimation of the domain of attraction. An illustrative example is given in order to show the efficiency of the proposed method
An agent-based Decision Support System for resources' scheduling in Emergency Supply Chains
No full textInternational audienceWe propose a multi-agent-based architecture for the management of Emergency Supply Chains (ESCs), in which each zone is controlled by an agent. A Decision Support System (DSS) states and solves, in a distributed way, the scheduling problem for the delivery of resources from the ESC supplying zones to the ESC crisis-affected areas. Thanks to the agents' cooperation, the DSS provides a scheduling plan that guarantees an effective response to emergencies. The approach is applied to two real cases: the Mali and the Japan crisis. Simulations are based on real data that have been validated by a team of logisticians from Airbus Defense and Space
A Distributed Frank-Wolfe Framework for Learning Low-Rank Matrices with the Trace Norm
No full textWe consider the problem of learning a high-dimensional but low-rank matrix from a large-scale dataset distributed over several machines, where low-rankness is enforced by a convex trace norm constraint. We propose DFW-Trace, a distributed Frank-Wolfe algorithm which leverages the low-rank structure of its updates to achieve efficiency in time, memory and communication usage. The step at the heart of DFW-Trace is solved approximately using a distributed version of the power method. We provide a theoretical analysis of the convergence of DFW-Trace, showing that we can ensure sublinear convergence in expectation to an optimal solution with few power iterations per epoch. We implement DFW-Trace in the Apache Spark distributed programming framework and validate the usefulness of our approach on synthetic and real data, including the ImageNet dataset with high-dimensional features extracted from a deep neural network
Isochoric heating and strong blast wave formation driven by fast electrons in solid-density targets
No full textInternational audienceWe experimentally investigate the fast (\atop\sim}10 ps into a 140 Mbar blast wave, according to hydrodynamic simulations, consistent with our measurements. These experimental and numerical findings pave the way to a short-pulse-laser-based platform dedicated to high-energy-density physics studies
Networking for ovarian rare tumors: a significant breakthrough improving disease management
No full textInternational audienceRare ovarian tumors represent >20% of all ovarian cancers. Given the rarity of these tumors, natural history, prognostic factors are not clearly identified. The extreme variability of patients (age, histological subtypes, stage) induces multiple and complex therapeutic strategies
Embedded Stem Priming Effects in Prefixed and Suffixed Pseudowords
No full textInternational audiencePrevious research has repeatedly revealed evidence for morpho-orthographic priming effects in suffixed words. However, evidence for the morphological chunking of prefixed words is sparse and ambiguous. The goal of the present study was to directly contrast the processing of prefixed and suffixed pseudowords within the same experiment. We carried out a masked primed lexical decision experiment, in which the same target (AMOUR [LOVE]) was preceded by a prefixed (preamour [prelove]), a nonprefixed (brosamour [broslove]), a suffixed (amouresse [lovedom]), and a nonsuffixed (amourugne [lovedel]) prime. The results revealed significant priming across all four conditions. Moreover, priming was modulated by individual differences in reading proficiency. High-proficiency readers showed evidence for embedded stem priming effects, independent of whether stems occurred in combination with a real affix or a nonaffix. This finding is of relevance to recent morphological processing theories, suggesting that embedded stems represent salient activation units during the reading of complex pseudowords
On the Troll-Trust Model for Edge Sign Prediction in Social Networks
No full textIn the problem of edge sign prediction, we are given a directed graph (representing a social network), and our task is to predict the binary labels of the edges (i.e., the positive or negative nature of the social relationships). Many successful heuristics for this problem are based on the troll-trust features, estimating at each node the fraction of outgoing and incoming positive/negative edges. We show that these heuristics can be understood, and rigorously analyzed, as approximators to the Bayes optimal classifier for a simple proba-bilistic model of the edge labels. We then show that the maximum likelihood estimator for this model approximately corresponds to the predictions of a Label Propagation algorithm run on a transformed version of the original social graph. Extensive experiments on a number of real-world datasets show that this algorithm is competitive against state-of-the-art classifiers in terms of both accuracy and scalability. Finally, we show that troll-trust features can also be used to derive online learning algorithms which have theoretical guarantees even when edges are adversarially labeled
On Graph Reconstruction via Empirical Risk Minimization: Fast Learning Rates and Scalability
No full textInternational audienceThe problem of predicting connections between a set of data points finds many applications, in systems biology and social network analysis among others. This paper focuses on the \textit{graph reconstruction} problem, where the prediction rule is obtained by minimizing the average error over all n(n-1)/2 possible pairs of the n nodes of a training graph. Our first contribution is to derive learning rates of order O(log n / n) for this problem, significantly improving upon the slow rates of order O(1/√n) established in the seminal work of Biau & Bleakley (2006). Strikingly, these fast rates are universal, in contrast to similar results known for other statistical learning problems (e.g., classification, density level set estimation, ranking, clustering) which require strong assumptions on the distribution of the data. Motivated by applications to large graphs, our second contribution deals with the computational complexity of graph reconstruction. Specifically, we investigate to which extent the learning rates can be preserved when replacing the empirical reconstruction risk by a computationally cheaper Monte-Carlo version, obtained by sampling with replacement B << n² pairs of nodes. Finally, we illustrate our theoretical results by numerical experiments on synthetic and real graphs
Augmented ETDs
No full textInternational audienceGraduate students have taken advantage of the various electronic theses and dissertations systems to enhance their capstone projects with supplemental files since the earliest days of ETDs. In 1997 when Timur Oral was required to submit his digital thesis to the Graduate School at Virginia Tech, " A Contemporary Turkish Coffeehouse Design Based on Historic Traditions, " it was augmented by two QuickTime movies. Along with the text, the PDF of his thesis contains many drawings and colorful pictures, but these now-primitive digital videos convey the atmosphere of a coffeehouse-everything but the coffee aroma. Recently the discussion of ETDs and their supplementary files has gained momentum. I