A0456
Title: A dynamic admixture Poisson process analysis of neuronal spike trains
Authors: Surya Tokdar - Duke University (United States) [presenting]
Abstract: The brain is able to encode multiple simultaneous stimuli and segment them into objects, but the neural computing behind this complex operation of great relevance to computational and cognitive neuroscience remains poorly understood. Presently lacking are statistical models and tools for quantifying how the response to a stimuli combination relates to the ensemble of activities evoked when each stimulus is presented independently. We seek solutions under an admixture theory that a single neurons response to multiple stimuli is a dynamically weighted average of its responses to individual items. We evaluated single unit activity in an auditory coding ``bottleneck'', the inferior colliculus, while monkeys reported the location(s) of one or two simultaneous sounds. Time-domain analyses of recorded spike trains were performed by assuming them to be realizations of inhomogeneous Poisson processes. Admixture behavior was modeled by using transformed Gaussian processes to capture dynamical averaging of single sound response rates. A Markov chain Monte Carlo algorithm was devised to perform Bayesian estimation of idiosyncratic trial level behavior as well as persistent cell level properties.
