Point process modeling on decoding and encoding for Brain Machine Interfaces

Point process modeling has the potential to capture the specificity of neural firing where the information is contained in the spike time occurrence. We aim at building an adaptive signal processing framework for brain machine interfaces working directly in the spike domain. However, the signal processing tools for continuous stochastic processes faces challenge when implemented directly on point processes. Under the Bayesian formulation, the effectiveness of the decoding algorithm and the accuracy of the encoding model will affect each other recursively on kinematics reconstruction. The finer time resolution of point process raises a higher computational complexity. This paper will review our recent work addressing these concerns. We implemented an instantaneous tuning model into a sequential Monte Carlo point process estimation to better reconstruct kinematics from neural spike trains for brain machine interfaces.