Alain de Cheveigne on cleaning up brain data for analysis and decoding

Techniques to measure brain activity (EEG, MEG, ECoG, etc.) all share a similar set of issues: high levels of noise, mixing between sources and observations, and limited dimensionality relative to the billions of sources active within the brain.  I will discuss approaches to address these issues. Source to sensor mixing is highly linear, and thus linear analysis methods are prominent, aiming to attempt to reverse the mixing or factor out major sources of noise and artifact. A wide range of techniques is available, including PCA, ICA, joint decorrelation, CCA, and others.  I will discuss my favorite methods, explain how they work and what they can do, and look at how we might overcome present-day limitations, to dig deeper into the haystack in search of the needle.

Bio: Alain de Cheveigné’s initial training was in Maths, Physics and EE.  He has worked on speech signal processing (pitch and spectral estimation), auditory modeling and psychophysics (pitch perception and sound segregation), and more recently, brain data analysis. He recently coordinated the European project COCOHA (cocoha.org) that aimed to develop a brain-controlled hearing aid. He works for CNRS and is affiliated with the Ecole Normale Superieure and University College London.