Synaptic Connectivity in Biologically Constrained Artificial Neural Networks

When: Wednesday, February 22, 2012 at 12:00 pm
Where: DA 114
Speaker: Julio Chapeton
Organization: Northeastern Unversity
Sponsor: Introduction to Physics Research

Nearly all cortical areas, from cerebellar to neocortical to hippocampal, share very similar features of circuit architecture. These areas are dominated by excitatory neurons and synapses, contain sparsely connected neural networks, and function with stereotypically distributed connection weights. Using the replica method from statistical physics, we show that such ubiquitous structural and functional features of cortical connectivity arise readily from the requirement of efficient associative memory storage. The theory makes two experimentally testable predictions. First, in spite of a large number of neuron classes, all excitatory to excitatory neuron connections between potentially connected cells must be realized with <50% connection probability. In contrast, all inhibitory to excitatory neuron connections must have >50% connection probability. Second, the coefficient of variation in connection weight must be uniquely related to the probability of that connection. These fundamental predictions do not depend on any parameters and must hold for any network optimized for learning.