Spiking Signal Modeling On Macaque's PMd And M1 Cortices Using Poisson Point Process

Neural coding and prediction is an active topic in neuroscience.By modeling on spiking signal to extract neural features,researchers can understand more of the brain's functionality as well as neural information delivering procedure across multiple cortices in cognitive processing,which would contribute to uncovering neuron's physiological property,even building neural prosthesis to achieve brain-machine fusion.This paper aims at building statistical model on macaque's spiking signals from PMd and M1 cortex to analyze their relationship.PMd and M1 cortex acts an important role in macaque's high-level cognitive activity.By modeling on these two cortices,researchers could extract more details of the collaboration between them.Neural modeling meets many challenges.For example,neural signals contain diverse properties related with neural cells,which requires model having strong expression to express the diversity.Besides,neural information is hidden in the point process array of spiking signals,which requires model to extract more features within this point process.We propose several enhancements based on Poisson-GLM to overcome these challenges,and our contribution is listed as follows:1.Inspired by the idea of blending model from ensemble learning,we train a series of sub-models with weak expression power,and blend these sub-models to enhance the expression ability of the whole model;2.We switch our target function of Poisson-GLM from maximizing likelihood function to optimizing Discrete Time Rescaling Kolmogorov Smirnov statistics,in order to enrich model's expression with regards to signal's point process feature;3.We examine our model's prediction from multiple aspects in our experiment,and the result shows our model maintains a remarkable result for goodness-of-fit evaluation,meanwhile the model maintains a good biological explanation.