Biophysics for Neural Computation

Summary

Date: 
October 6, 2020 - 12:00pm
Location: 
Bio Labs 1080
About the Speaker
Name: 
Mark Harnett
Speaker Title: 
Assistant Professor
Speaker Affiliation: 
MIT

The biological substrates of computation in the cortex remain elusive. While the remarkable recent progress of artificial neural systems has generated important insights into cortical function, these networks employ highly simplified model neurons. Real cortical neurons receive thousands of synaptic inputs distributed across extensive dendritic arbors that exhibit highly nonlinear properties, providing an opportunity for subcellular computation before final integration and output at the axon. Neuronal units endowed with nonlinear dendritic processing could provide their respective networks with increased power, flexibility, and/or efficiency. However, the contributions of dendrites to cortical computations underlying behavior remain unclear. Dendritic mechanisms may not be recruited during relevant in vivo circuit activity or may only serve to compensate for other biological constraints. In this talk, I will discuss my lab’s recent progress in evaluating the biophysical substrates, engagement, and utility of dendritic processing in the mammalian cortex. I will first describe a multidisciplinary synapses-to-systems analysis of the mouse retrosplenial cortex that aims to connect biophysical processes at the level of single neurons to the computations carried out by cortical populations during ethologically relevant navigational behavior. I will then discuss our systematic efforts to understand how biophysical properties (ion channels and dendritic morphology) shape neuronal input-output functions for a single identified cell type (L5b pyramids) across the phylogenetic tree of mammals, from Etruscan shrews to humans. We hope to use the results from these two lines of inquiry to understand general principles of cortical design and function, and to facilitate the development of new artificial neural networks with enhanced capabilities.