Orchestrated Excitatory and Inhibitory Learning Rules Lead to the Unsupervised Emergence of Self-sustained and Inhibition-stabilized Dynamics

Abstract

Self-sustaining dynamics maintained through recurrent connections are of fundamental importance to cortical function. We show that Up-states—an example of self-sustained network dynamics—autonomously emerge in cortical circuits across three weeks of ex vivo development, establishing the presence of unsupervised synaptic learning rules that lead to globally stable emergent dynamics. Computational models of excitatory-inhibitory networks have established that four sets of weights (WE←E, WE←I, WI←E, WI←I) cooperate to generate stable self-sustained dynamics, but have not addressed how a family of learning rules can operate in parallel at all four weight classes to achieve self-sustained inhibition-stabilized regimes. Using numerical and analytical methods we show that standard homeostatic rules cannot account for the emergence of self-sustained dynamics due to the paradoxical effect. We derived a novel family of homeostatic learning rules that operate in parallel at all four synaptic classes, which robustly lead to the emergence of Up-states and balanced excitation-inhibition.

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

“…In more recent recurrent network models, only a fraction of excitatory and inhibitory synapse-types are modeled as plastic, and neural responses exhibit a narrow subset of the different response patterns recorded in experiments. 14,24 – 29…”

“…In more recent recurrent network models, only a fraction of excitatory and inhibitory synapse-types are modeled as plastic, and neural responses exhibit a narrow subset of the different response patterns recorded in experiments. 14,[24][25][26][27][28][29] Here we present a Hebbian learning framework with minimal assumptions that explains a wide range of experimental observations. In our framework, synaptic strengths evolve according to a Hebbian plasticity rule that is stabilized by the competition for a limited supply of synaptic resources.…”