This notebook is the work produced by Omar Ahmad and Matias Etcheverry, in the course Apprentissage pour les séries temporelles given by Laurent Oudre and Charles Truong at the MVA.

The goal is to study the notion introduced by the article Multivariate Temporal Dictionary Learning for EEG. The aim of this article is to generalize the concept of dictionary learning to multivariate signals with shift-invariant atoms.

We proposed to study this algorithm implemented on this repository. We focus on the dataset 2a of the Brain Computer Interface IV competition (available here and in the data folder).

Table of content:

1. Dictionary Learning
2. Dictionary learning in a noised setting
3. Dictionary learning with different initializations
4. Classification from activations