High Performance Optimization of Independent Component Analysis Algorithm for EEG Data

Independent Component Analysis (ICA) is known as a signal cleaning method that allows the artifacts to be extracted and subsequently eliminated. It is especially essential while processing the EEG data. However, this is a time-consuming algorithm especially if we deal with a high-dimensional data and take care about the calculation accuracy. One of the known implementations of this algorithm, which can be found in MATLAB or the open library it++- fastICA -does not use parallel implementations nor take benefit of the current capabilities of the Intel architecture. Also for large data, fastICA's accuracy and stability decrease due to the reduction of data dimension. The paper introduces an implementation that uses Intel Cilk Plus, BLAS and MKL library built-in functions as well as array notation and OpenMP parallelization to optimize the algorithm.