ausblenden:
Schlagwörter:
-
Zusammenfassung:
Extending on previous work, the Echo State Network (ESN) and Tikhonov Regularisation (TR) training algorithms were implemented for both the CPU, an Intel i7-980; and the GPU, an Nvidia GTX480. The implementation used all 4 cores of the CPU, and all 480 cores of the GPU. The execution times of these implementations were measured and compared. In the ESN case, speed-ups were observed at reservoir sizes greater than 1,024. The first significant speed-ups of 6 and and 5 were observed at a reservoir size of 2,048 in double and single precision respectively. In the case of Tikhonov Regularisation, no significant speed-ups were observed.
