Point process regression models of neural synchrony
This talk covers statistical analysis of spike train data, the modeling approach GLM, and the problem of assessing neural synchrony.
Talk abstract: Over roughly the past 10 years point process regression models, often called generalized linear models or GLMs, have become a standard tool for relating neural spiking activity to putative causes, such as features of stimuli, network dynamics, and intrinsic dynamics of neurons. I will begin by discussing the virtues of this modeling approach, and will then turn to the problem of assessing neural synchrony. Synchrony is widely believed to play a fundamental role in neural computation, but its statistical assessment is subtle. I will describe how point process regression models generalize what are commonly called maximum entropy models, and can accommodate the kind of time-varying firing rates that typically appear in stimulus-driven (or action-driven) neurons, as well as non-Poisson or network effects. One of the problems faced in large array recordings is the statistical control of false discoveries. A Bayesian method for controlling false discoveries has yielded interesting physiological results from Utah array recordings in primary visual cortex.