MASSIVELY PARALLEL NEURAL SIGNAL PROCESSING

Abstract

This thesis investigates the processing of neural signals. Spike detection has a profound impact on neural decoding accuracy. However, most existing spike detection algorithms are supervised and require human assistance. There is also no objective way to compare the performance of different algorithms due to the lack of ground truth in real data. Using a monkey BCI experiment, we compared the performance of EC-PC, a non-supervised spike detection algorithm previously developed in our group with other algorithms and found that EC-PC produced the highest decoding accuracy. Large-scale neural recordings are necessary to understand the neural mechanisms of the brain due to its complexity. However, the computational burden to process this large amount of data is huge. We have created a massively parallel neural signal processing software package in GPU to process data efficiently. The GPU package offers 5x to 110x speedup compared to a CPU program.