Recherche et sélection de publications
Interface en ou

Caveats with stochastic gradient and maximum likelihood based ICA for EEG

Jair Montoya #1, Jean-François Cardoso #1, Alexandre Gramfort #1
#1 Télécom ParisTech
  • Institut Mines-Télécom
References
LVA/ICA, Grenoble, February 2017,
Abstract

Stochastic gradient (SG) is the most commonly used optimization technique for maximum likelihood based approaches to independent component analysis (ICA). It is in particular the default solver in public implementations of Infomax and variants. Motivated by experimental findings on electroencephalog- raphy (EEG) data, we report some caveats which can impact the results and inter- pretation of neuroscience findings. We investigate issues raised by controlling the step size in gradient updates combined with early stopping conditions, as well as initialization choices which can artificially generate biologically plausible brain sources, so called dipolar sources. We provide experimental evidence that push- ing the convergence of Infomax using non stochastic solvers can reduce the num- ber of highly dipolar components and provide a mathematical explanation of this fact. Results are presented on public EEG data.

Keywords
Independent component analysis (ICA), maximum likelihood, stochas- tic gradient method, infomax, electroencephalography (EEG), neuroscience
Category
Paper in proceedings
Research Area(s)
Computer Science/Signal and Image Processing
Identifier(s)
Bibliographic key jmontoyaicalva2017
Export
Last update
on february 01, 2017 by Jair Montoya Martinez


Responsable du service
Dominique Asselineau dominique.asselineau@telecom-paristech.fr
Copyright © 1998-2017, Télécom ParisTech/Dominique Asselineau