Spikes, synapses, learning and memory in simple networks
Surya Ganguli

Multiple time scales underly neural computation, coding and dynamics. The interactions between biophysical dynamics occuring on these multiple time scales presents a significant challenge in understanding their collective contribution to neural computation and coding. For example how might fast spiking dynamics, intermediate range oscillations, and slower changes in synaptic efficacies collectively give rise to even longer lasting effects, such as long term memory traces? We will explore this type of question in simple model neural networks. We will first show how to explicitly compute the statistical properties of multineuronal spike trains directly in terms of such networks' synaptic efficacies and latencies. We then apply this result to generate and analyze an effective dynamics on the space of synaptic parameters that captures the slow evolution of synaptic patterns in response to spiking inputs. Such an effective dynamics can aid in bridging the time scales between fast spiking dynamics and slower changes in long term memory traces due to changes in synaptic patterns. Finally we comment on how simple models of this form might also aid in solving an inverse problem: given multineuronal spiking statistics, how can one infer underlying synaptic efficacies and latencies?