-----------------------------------------------------------

Applied Mathematics Colloquium

Friday, September 14, 11:30 am
Cullimore Lecture Hall II
New Jersey Institute of Technology

Feedforward and Recurrent Processing in Cat Primary Visual Cortex

Ken Miller

Center for Theoretical Neuroscience

Columbia University

New York, NY

Abstract

Cat primary visual cortex (V1) has served for many years as a model system for understanding processing in cerebral cortex. The cortex shows strong recurrent excitation and feedback inhibition, and theories of visual selectivity have been based on this. Yet much work -- including theoretical work from my group and experimental work from other groups -- shows that the major features of visual selectivity of V1 neurons can be accounted for by feedforward processing, through a combination of tuned excitatory feedforward drive, untuned inhibitory feedforward drive, and intrinsic cellular and synaptic nonlinearities. What then is the role of the enormous recurrence? Our recent work provides some hints. First, in collaboration with David Ferster's lab, we have studied surround suppression, in which stimuli outside the classical receptive field (cRF: the visual region that can drive a cell's responses) can modulate responses to cRF stimuli. We find that the data on surround suppression suggest V1 functions as an inhibition-stabilized network (ISN), in which excitatory recurrence alone is strong enough to be unstable but is stabilized by feedback inhibition. Second, we find that the stable combination of strong recurrent excitation and feedback inhibition leads to large and spatially structured amplification of inputs without a network elongation of the timescales of activity, a phenomenon we call "transient amplification". Biologically, this suggests an explanation for observations of structure in spontaneous V1 activity; mathematically, this derives from the strong non-normality of the matrices describing synaptic connectivity. The ISN is a structure that allows stable integration of excitatory input from many sources. This along with the apparently feedforward nature of primary selectivity suggests that the strong cortical recurrence may be a specialization to allow strong contextual modulation -- strong interaction between local regions that each reflect their own feedforward input.