Towards mathematically tractable, normative models of neural computation
Michael Buice, Ph.D.
Associate Investigator, Allen Institute for Brain Science
Department of Applied Mathematics,
University of Washington
Seminar abstract: To achieve even a partial understanding of neural computation we must 1) define mathematically tractable descriptions of complex systems that permit definitively testable predictions and 2) relate these descriptions to measurements of real neural systems in a way that permits the comparison of competing models. I will describe first a mathematically tractable approach to developing simplified descriptions of neural systems based on their constituents. I will focus in particular on applying these methods to systems with partial measurements, in which only a fraction of neurons is measured, as is typical in systems neuroscience. I will then describe the use of the Allen Brain Observatory, a large-scale physiological data set, as a test bed for systems-level normative models of neural computation, namely task-trained artificial neural networks. Finally, I will argue for an approach to modeling neural systems that incorporates both the mathematical properties of models and the experimentally available variables.
Mechanisms underlying flexible information flow across the brain Karel Svoboda, Ph.D. Director, Allen Institute: Abstract: Neural computation and behavior are produced by shifting configurations of multi-regional neural networks, implemented by dynamic coupling between brain regions. We...