BMI SEMINAR // Samuel Muscinelli - Firing rate adaptation shapes intrinsic fluctuations and signal transmission in recurrent neural networks

Cortical neurons exhibit multiple history-dependent mechanisms such as refractoriness and spike-frequency adaptation. Neurons are embedded in highly recurrent networks, and the recurrent feedback is believed to play a key role in the generation of the irregular fluctuations that are observed in neuronal recordings. However, an understanding of how single-neuron mechanisms, such as adaptation, interact with recurrent connectivity to shape the network dynamics is largely lacking.
We study the effect of adaptation on the dynamics of large random neural network models using techniques derived from statistical physics. We find that the introduction of adaptation shifts the network to a new dynamical regime, in which the fluctuations while remaining chaotic, are dominated by a resonance frequency. This new regime has dramatic consequences on the way the network responds to external signals, as the introduction of adaptation strongly favors the response to low-frequency signals.