2019/11/11 by Chris Glynn, Surya T. Tokdar, Glynn, Chris +11
Agricultural and Biological Sciences · #Animal Nutrition and Physiology #Applications (stat.AP) #FOS: Computer and information sciences
paper · pdf · doi:10.48550/arxiv.1911.04387
openalex publication_date 2019/11/11 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Conventional analysis of neuroscience data involves computing average neural\nactivity over a group of trials and/or a period of time. This approach may be\nparticularly problematic when assessing the response patterns of neurons to\nmore than one simultaneously presented stimulus. In such cases, the brain must\nrepresent each individual component of the stimuli bundle, but\ntrial-and-time-pooled averaging methods are fundamentally unequipped to address\nthe means by which multi-item representation occurs. We introduce and\ninvestigate a novel statistical analysis framework that relates the firing\npattern of a single cell, exposed to a stimuli bundle, to the ensemble of its\nfiring patterns under each constituent stimulus. Existing statistical tools\nfocus on what may be called "first order stochasticity" in trial-to-trial\nvariation in the form of unstructured noise around a fixed firing rate curve\nassociated with a given stimulus. Our analysis is based upon the theoretical\npremise that exposure to a stimuli bundle induces additional stochasticity in\nthe cell's response pattern, in the form of a stochastically varying\nrecombination of its single stimulus firing rate curves. We discuss challenges\nto statistical estimation of such "second order stochasticity" and address them\nwith a novel dynamic admixture Poisson process (DAPP) model. DAPP is a\nhierarchical point process model that decomposes second order stochasticity\ninto a Gaussian stochastic process and a random vector of interpretable\nfeatures, and, facilitates borrowing of information on the latter across\nrepeated trials through latent clustering. We present empirical evidence of the\nutility of the DAPP analysis with synthetic and real neural recordings.\n