Spatio-temporal Irregularity and Multi-stability in Balanced Networks with Short-term Synaptic Plasticity
Joint work with: Carl van Vreeswijk and David Hansel.
In physiological conditions, cortical dynamics exhibits significant spatio-temporal irregularity. Spike trains are characterized by high coefficient of variations and exponential-like distributions of inter-spike intervals. Distributions of average emission rates are right skewed and long tailed, with rates ranging from below 1Hz to several tens of Hz. This pattern of activity does not depend on the cortical region, on the behavioral state nor on the conditions of activation of the network. Theoretical studies, supported by experimental results, indicates that spatio-temporal irregularity is a signature of a basic mode of operation of local cortical circuits: dynamical balance of excitation and inhibition. Theory also predicts, however, that networks operating in such a regime would respond linearly to external inputs. This poses a serious problem for neurophysiologically-based models of computation for, in those models, computation relies to a large extent on non-linearities. The problem is especially severe for the attractor framework, where computation relies on the network being able to exhibit multiple, steady states of activity, and within which important computations, as short-term memory and decision making, have been modeled. By which mechanisms could non-linearity and multi-stability be restored in networks working in the balanced regime? We have investigated the possibility that non-linearity in the balanced regime be provided by short-term synaptic plasticity (STP). We have shown that balanced networks with STP do indeed exhibit a bi-stable regime, where a non-zero, low rate state coexists with an high-rate state of activity. In both states, the patterns of network activity exhibit significant spatio-temporal irregularity, consistently with experiment. In this scenario, bi-stability is achieved by a change in the levels of (self-generated) synaptic noise, rather than by changes in the mean drive. Importantly, the bi-stable regime achieved by this mechanism is tolerant to significant levels of spatio-temporal irregularity in the patterns of network activity. These results highlight a new functional role for STP, and lay the groundwork for a principled understanding of functional/computational implications of the balanced regime.