high performance computing on graphics processing units: hgpu.org

Sequential Monte Carlo estimation on point processes has been successfully applied to predict the movement from neural activity. However, there exist some issues along with this method such as the too simplified tuning model and the high computational complexity. In this paper, we attempt to address these issues and improve its decoding performance. Firstly, a more general tuning model which takes the recent ensemble activity into account is proposed. The goodness of fit analysis demonstrates that the model can predict the neuronal response more accurately than the one only depends on kinematics. Then a new sequential Monte Carlo algorithm based on the proposed model is constructed, and the decoding accuracy is evaluated using root mean square error across 8 datasets. The results show that our proposed algorithm can significantly reduce the decoding error, which decreases about 23.6% in position, compared to the one only depends on kinematics. In addition, we accelerate the decoding speed by implementing the algorithm in massive parallel and run it on GPU. The experimental results demonstrate that compared with the serial implementation, our parallel method runs over 10 times faster. It could decode the neural activity in real time (<10ms) even with thousands of particles or hundreds of neurons. The main contribution of our work is to enable the sequential Monte Carlo algorithm with point process observation to output the estimation much faster and more accurate, which is greatly helpful for the development of high-performance Brain Machine Interfaces.