Control of recurrent neural network dynamics byhomeostatic intrinsic plasticity

International audienceHomeostatic intrinsic plasticity (HIP) is a ubiquitous cellular mechanism regulating neuronalactivity, cardinal for the proper functioning of nervous systems. Here,we assess how cellular HIP effects translate into collective dynamics and computationalproperties in biological recurrent networks. We develop a realistic multi-scale modelincluding a generic HIP rule regulating the neuronal threshold with actual molecular signalingpathways kinetics, Dale’s principle, sparse connectivity, synaptic balance and Hebbiansynaptic plasticity (SP). Dynamic mean-field analysis and simulations unravel that HIP sets aworking point at which inputs are transduced by large derivative ranges of the transferfunction. This cellular mechanism insures increased network dynamics complexity, robustbalance with SP at the edge of chaos, and improved input separability. Although criticallydependent upon balanced excitatory and inhibitory drives, these effects display strikingrobustness to changes in network architecture, learning rates and input features. Thus, themechanism we unveil might represent a ubiquitous cellular basis for complex dynamics inneural networks. Understanding this robustness is an important challenge to unravel principlesunderlying self-organization around criticality in biological recurrent neural networks