Logo Supelec

The SSE department from Supelec has been working for the last decade on a research theme named TSI (Traitement Statistique de l’Information - Statistical processing of information). Within this framework, it has acquired a renowned expertise on black-box-type modelisation problems as well as uncertain information management. These methods include random processes linear prediction (kriging) as well as regularized regression by reproducing kernel hilbertian-space norm(splines, approximation by radial basis functions, support vector regression, etc. )

The most recent work has been on 1/ kernel choice, a central problem since model quality is dependant upon it. Methods assembling a kernel from many elementary kernels are being studied. They allow satisfying results, in particular for time series predictions. 2/ continuous time dynamic systems prediction usually use a discrete form for time. An original method allowing processing of the modelisation problem without system discretization has been proposed. 3/ the approximation phase of a costly function, the global optimum of which we are seeking, can be done using these methods. In the past, the methods used tried to optimize the approximation without taking the uncertainty into account, or were optimizing the expected gain by taking uncertainty into account but ignoring the global information gain from new observations. We present an optimization method based on entropy reduction that allows conciliation of both methods' advantages. 4/ tail behaviour and system failure probability can be processed by a gaussian process' excursion volume estimation.

SSE department is strongly involved in :

· DIGITEO-Labs (RTRA Digiteo) skill pole, especially in Incerteo axis (research about uncertainty definition, modelisation and propagation) ;

· Num@tec Automotive and Usine Numérique projects from the SYSTEM@TIC PARIS-REGION competitivity pole;

· IMdR, Redopt, GdR MASCOT-NUM, ISIS workgroups (among others).

SUPELEC website.

Last Updated (Monday, 11 January 2010 22:37)