UVF4B101: Machine Learning - Fall 2016/2017

FIP 3A IR Automne
MR 2A SISEA-I Automne
MR 2A SISEA-S Automne
MR 2A MARS Automne
IG 3A Automne
The theory of statistical decision lies at the heart of signal processing. This course covers notions of detection, estimation and classification. The usual decision criteria are presented (maximum likelihood, a posteriori maximum, mean squared error), then applied in the case of Gaussian signals. EM and MCMC methods are tackled. Mean squared linear estimation plays an important role in processing methods, leading to the notion of optimal filtering. Two cases are examined: the Wienernon recursive numerical filter, used for stationary signals, and the Kalman recursive numerical filter, which can be extended to non stationary signals. Specific filtering, applied to non Gaussien noise linear systems is introduced. Another modeling, relying on hidden discrete state Markov models is examined, leading on to the Baum-Welch estimation and a contextual classification using the Viterbi algorithm The unit also addresses methods for analysing the behaviour of non stationary random signals simultaneously in time and frequency. The short term Fourier transform is the method which historically was the first used. The study of energy distribution in time and frequencies allows the definition of a real Time-Frequency transformation, called Wigner-Ville. The Time-Scale analysis has renewed this vision via the use of continuous and discrete Wavelets extending to a multiresolution analysis of the signal. Statistical filtering may be developed according to this principle, as well as compression methods.
Location  : BREST
Coordinator  : Thierry CHONAVEL
Credits FIP 3A IR  : 6
Credits IG 3A  : 6
Dernière màj le 24-OCT-16 par RETIF
Code Intitulé
Co-resp. Etat
Date màj
Last update
F4B101A Traitement statistique avancé de l'information T.Chonavel D.Pastor Validée 07-03-16

IMT Atlantique
Campus de Brest
Technopôle Brest-Iroise
CS 83818
29238 Brest Cedex 3

Tél  +33 (0)2 29 00 11 11
Fax +33 (0)2 29 00 10 00