Independent component analysis (ICA) is a widely used data exploration technique in neuroscience, it is part of most EEG/MEG processing pipelines. It aims at decomposing signals into a mixture of independent sources.

The most common ICA solvers, FastICA and Infomax, converge slowly when applied to real data.

 

Faster ICA: PICARD

 

Picard is a new ICA solver. It solves the exact same problems as Infomax and FastICA, much faster on real signals. We get the following convergence curves on EEG datasets of 71 channels with ~ 300K time samples:Open-source Python and Matlab code is available online: https://pierreablin.github.io/picard/

References

Pierre Ablin, Jean-François Cardoso, Alexandre Gramfort
Faster Independent Component Analysis by Preconditioning with Hessian Approximations,
IEEE Transactions on Signal Processing, 2018

Pierre Ablin, Jean-François Cardoso, Alexandre Gramfort
Faster ICA under Orthogonal Constraint,
ICASSP, 2018