CTN Seminar: Gianluigi Mongillo: The balance of excitation and inhibition and robust, high-capacity memory networks
According to a major theoretical framework, memories are retrieved as stable steady states of the network dynamics. We investigate the conditions that allow neuronal networks to possess a large number of such steady states, that is, to exhibit large memory capacity. We find that large memory capacity requires a very precise cancellation of the excitatory and inhibitory drive at the single-neuron level. This cancellation can be achieved either structurally - the total excitatory and inhibitory synaptic efficacies are the same - or dynamically - the total excitatory and inhibitory synaptic inputs are the same. In both cases, the network displays optimal memory capacity. However, only when the cancellation is achieved dynamically, memory retrieval is robust to large fluctuations in the average level of neuronal activity and/or in the synaptic connectivity. Our results provide a computational rationale for the dynamical balance of excitation and inhibition as observed in the cortex.