GPU optimization of EEG analysis

RAIMONDO, F; KAMIENKOWSKI JE; FERNANDEZ SLEZAK, D

Córdoba

Conferencia; 40 JAIIO; 2011

Nowadays, with the advent of new non-invasive techniquesof brain imaging, researchers have access to neural processes underlyingthe cognition in humans. One of the main challenges in this techniques isthe detection of patterns in brain signals, generally very noisy and withartifacts inserted by vital signs. One of the most successful techniquesfor this is Independent Component Analysis which detects statisticallyindependent components that are produced from dierent sources. Thesemethods are very expensive in computational time, with many hours ofprocessing for a single experiment. We analyzed this algorithm and detecttwo main types of operations: vector-matrix and matrix-matrix. Weimplemented an ad-hoc solution that executes on GPU and comparedthis with the original and CUBLAS versions. We obtained a 4x and 40xof performance increase of vector-matrix and matrix-matrix operations,respectively. These results are the rst step towards real-time EEG processingwhich may produce a signicant advance into BCI applications.