Université de technologie de Troyes

 

Contenu

Séminaires programmés

Mardi 7 Mars 2017

Elham Mosayeb vient de l’Université de Téhéran. Elle est en séjour parmi nous pendant le semestre en cours. Elle fera mardi prochain (7 mars) de 14h à 15h  une présentation de ses travaux de recherche en salle G110.

Son exposé portera sur «  Optimal Accelerated Degradation Test Plan Based on Inverse Gaussian Process »  

Venez écouter Elham et discuter avec elle de son sujet de recherche dont voici le résumé.

Using stochastic processes is a common way to model the degradation process. Wiener and Gamma processes are the most frequent used stochastic processes in this issue. In spite of their good properties, these two models can not handle modeling all degradation data. For example, it is been found that neither models fits the GaAs laser degradation data well. Another attractive degradation model which is recently introduced is inverse Gaussian (IG) process. Similar to Gamma process, it has monotone path and it has been shown to be a limiting compound Poisson process, which gives it a meaningful physical interpretation for modeling degradation of products deteriorating in random environments. Moreover, it has been shown that compared with Gamma process it has superb properties when dealing with covariate and random effects. Also the Bayes inference can be done with no difficulty or effort because of its close relation with the Wiener process with drift. Here, our target is to discuss about accelerated degradation test (ADT) planning for the IG process. Constant accelerated degradation test (CSADT) and step stress accelerated (SSADT) as two kinds of ADT are considered and under the constraint that the total experimental cost does not exceed a predetermined budget the optimal tests are obtained.

 

Séminaires passés

Jeudi 26 Mars 2015

Diego Rodolpho Tomassi, Chercheur à l'IMAL/ Argentine (Instito de Matematica Aplicada del Litoral), checheur invité à l'UTT.

Dimension reduction using measures of statistical dependence.Reducing the dimensionality of the features is a common preprocessing step for classification problems. Sufficient dimension reduction is a methodology that attempts to preserve all the information available in the original data that is relevant for a particular objective. Under this approach, estimation is usually cast into an inverse regression framework, requiring some assumptions on the data distribution. In this work we present a semi-parametric approach to estimate the sufficient reduction, driven by measures of statistical dependence proposed recently in the literature. An extension to include simultaneous variable selection is also discussed. This is an ongoing research collaboration with Pierre Beauseroy of UTT. 

Jeudi 12 Mars 2015

Olga Klopp, MdC Paris Ouest Nanterre

Complétion de matrices 1-bit

Jeudi 22 Janvier 2015

Elodie Chantery, MdC INSA Toulouse

Diagnostic et Prognostic des systèmes hybrides

Jeudi 27 Novembre 2014

Yingjun Deng, doctorant LM2S,

A piecewise quasi-linear Monte-Carlo method for thé first passage problem of an Ornstein-Uhlenbeck process

Jeudi 19 Juin 2014

Hai Canh Vu, doctorant LM2S,

Regroupement dynamique des systèmes de structure quelconque

 Jeudi 15 mai 2014

Nelly Pustelnik,  chargée de recherche CNRS à ENS-Lyon

Empirical Mode Decomposition revisited by multicomponent non smooth convex optimization

 Jeudi 13 mars 2014

 Charanpal Dhanjal,  chercheur Postdoc à Télécom ParisTech.

Efficient Eiggen-updating for Spectral Graph Clusterin

 Jeudi 20 février

 Gieuseppe Valenzise, chercheur CNRS à Télécom Paris-Tech/ LTCI.

High dynamic range video coding: challenges and emerging approaches

Mercredi 5 février à 14h

Nataliya Sokolovska, MCF à l'Université Pierre-et-Marie-Curie  à l'équipe bio-informatique.

Efficient training of sparse conditional random fields