Porovnávání EEG signálů s využitím GPU

Abstract

The aim of this bachelor thesis is the exploration of methods of EEG signals comparison, as well as the implementation of chosen method using CUDA enabled GPU to accelerate the chosen algorithm. In the first part, the theory of pattern recognition and signal processing is described. BCI systems exploiting EEG signals are discussed and the state of the art methods for EEG signal comparison are briefly introduced. The second part deals with NMF problem, the chosen NMF method -- FastHALS algorithm -- is described. The third part describes design and implementation of parallel solution of algorithm with the use of CUDA enabled GPU. The fourth chapter consists of performance tests of proposed solution. The conclusion contains evaluation of results. Possible extensions and improvements are discussed.