Recurrent Spiking Neural Networks for Continuous Motor Control: A Biologically Plausible Model of Motor Cortex

A recurrent spiking neural network model of motor cortex learns continuous motor control using biologically local learning signals, including Feedback-Spike Nudging delivered as spikes.

• Why it matters: It addresses how motor cortex can acquire skilled continuous control through recurrent spiking and experience-dependent plasticity without relying on biologically implausible learning signals.
• What they did: The model derives recurrent firing-rate state from synaptic input and trains via implicit differentiation, then uses State-Carry Initialization for closed-loop control and FSN for spike-based feedback learning.
• The result: Networks learn continuous control with supervised and reinforcement learning, with FSN consistent with mouse motor cortex spine plasticity and potentially mapping to cerebellum and basal ganglia interactions.

AI summary of the abstract · may contain errors · not medical advice · read the paper.