Developing a Data Driven System for Computational Neuroscience

Abstract. A data driven system implies the need to integrate data ac-quisition and signal processing into the same system that will interact with this information. This can be done with general purpose proces-sors (PCs), digital signal processors (DSPs), or more recently with field programmable gate arrays (FPGAs). In a computational neuroscience system that will interact with neural data recorded in real-time, clas-sifying action potentials, commonly referred to as spike sorting, is an important step in this process. A comparison was made between using a PC, DSPs, and FPGAs to train a spike sorting system using Gaussian Mixture Models. The results show that FPGAs can significantly outper-formed PCs or DSPs by embedding algorithms directly in hardware.