......................................................................

NJIT Mathematical Biology Seminar

Tuesday, March 31, 2009, 4:00pm
Cullimore Hall 611
New Jersey Institute of Technology

......................................................................

Nonparametrics for spike train analysis

Asohan Amarasingham

Center for Molecular and Behavioral Neuroscience, Rutgers University

Large-scale, multi-neuronal spike train recordings have now become routine. These records represent observations on the activity of a number of units embedded within a far larger and complex system, the dynamics of which are poorly-understood. Accordingly, they present a significant challenge to off-the-shelf, parametric statistical techniques, wherein the contribution of model misspecification to inferential error is arguably difficult to judge, or verify. We will argue that this makes the case for specially-crafted, nonparametric, and exploratory statistical tools. We will discuss two such classes of techniques. First, we will examine a few instances of a broad class of techniques based on the jitter principle, wherein a "jitter ensemble" of temporally perturbed versions of recorded spike trains are studied to infer their temporal structure. Second, we will discuss a resampling-based multiple hypothesis testing technique for inferring differences in neuronal firing rates, when these differences are instantiated across varying experimental conditions and multiple time scales. Conditional inference provides a unifying principle. (This is joint work with Gyorgy Buzsaki, Matthew Harrison, Stuart Geman, and Shigeyoshi Fujisawa.)