EEG segmentation through time­-varying PCA

Maxime Rio (1) , Axel Hutt (1) , Bernard Girau (1)

1 CORTEX - Neuromimetic intelligence

In this study, we have investigated a new method to localize ERPs components inside EEG records, in time and space. This method relies on a probabilistic model using both hidden Markov model and principal components analysis. In order to automatically select the model dimension, a Bayesian approach has been employed. Inference is realized by a variational approximation.

Mots clés: Evoked related potentials; principal components analysis; hidden Markov model; variational inference

Domaines: Analyse numérique [cs.NA]; Machine Learning [stat.ML]

Maxime Rio, Axel Hutt, Bernard Girau. EEG segmentation through time­-varying PCA. Neurocomp 2009, Sep 2009, Bordeaux, France. p. P-28. ⟨hal-00402047⟩