From Local Learning Rules to Systems of Nonlinear Volterra Integrodifferential Equations
Michael Wu, University of California, Berkeley
Using what we know from visual neurophysiology, we are able to write down a system of nonlinear integrodifferential equations that governs the development of the visual cortex. We’ve discretized and solved the system numerically with the forward Euler method. Although suboptimal, we are able to observe symmetry breaking and redistribution of lateral weights. This system has flexible features that enable us to incorporate many biological details without much more computational load. Specifically, we will discuss ways to incorporate synaptic efficacy and multi-time scale plasticity into our system. Finally, we will briefly discuss the optimization of numerical methods for solving this system and possibilities of simplifying the system without sacrificing its essential features.
Abstract Author(s): Michael C. K. Wu