Faster ICA under orthogonal constraint

Pierre Ablin (1, 2) , Jean-François Cardoso (3, 4) , Alexandre Gramfort (1, 2)

1 PARIETAL - Modelling brain structure, function and variability based on high-field MRI data
2 Université Paris-Saclay
3 IAP - Institut d'Astrophysique de Paris
4 CNRS - Centre National de la Recherche Scientifique

Independent Component Analysis (ICA) is a technique for unsupervised exploration of multi-channel data widely used in observational sciences. In its classical form, ICA relies on modeling the data as a linear mixture of non-Gaussian independent sources. The problem can be seen as a likelihood maximization problem. We introduce Picard-O, a preconditioned L-BFGS strategy over the set of orthogonal matrices, which can quickly separate both super-and sub-Gaussian signals. It returns the same set of sources as the widely used FastICA algorithm. Through numerical experiments, we show that our method is faster and more robust than FastICA on real data.

Mots clés: Independent component analysis; blind source separation; quasi-Newton methods; maximum likelihood estimation; preconditioning

Domaines: Machine Learning [stat.ML]; Applications [stat.AP]

Pierre Ablin, Jean-François Cardoso, Alexandre Gramfort. Faster ICA under orthogonal constraint. International Conference on Acoustics, Speech, & Signal Processing, 2018, Calgary, Canada. ⟨hal-01651842⟩