Penalized Regression Methods in the Source Analysis of Face Recognition

Recent developments in the field of variable selection through penalized least squares regression provide means for the analysis of neuroscience data. Particularly, combinations of non-convex penalties allow for sparse solutions and other unexplored properties that are especially attractive in their application to e.g. EEG/MEG inverse problem. Here, we explore the use of these techniques for the source analysis of a cognitive process, namely, the recognition of faces. Found sources are in agreement with previous studies and new methods, based on combination of penalties, provided for more physiologically plausible solutions.