Local homeostatic scaling supports stable rate propagation under noise and heterogeneity

Abstract

Reliable transmission of firing-rate signals through neural circuits requires synaptic coupling to remain within a narrow regime shaped by neuronal heterogeneity and noise. Outside this regime, classical theory predicts that activity will dissipate or amplify into synchrony. Studies have shown that stable rate-code transmission is biologically achievable, but they leave open how the required operating regime is established. Canonical plasticity rules including Hebbian plasticity and conventional homeostatic scaling do not by themselves recover this regime. We therefore introduce relational homeostatic scaling, a local synaptic scaling rule in which each postsynaptic neuron adjusts its total excitatory afferent strength to reduce the discrepancy between a recent AMPA-weighted afferent activity trace and a recent postsynaptic activity trace. Together, these traces capture the recent rate entering and leaving the neuron without requiring a global rate target. Mean-field analysis and simulations show that this rule drives synaptic weights toward the critical regime without precise weight initialization. The rule remains robust to neuronal heterogeneity, composes with spike-timing-dependent plasticity without disrupting Hebbian competition, and stabilizes the eligibility trace required by three-factor learning rules. These findings suggest that relational homeostatic scaling may provide a substrate for stable rate-code transmission in neural circuits. Author SummaryNeural signals often pass through many processing stages, but this is difficult in networks made of noisy and heterogeneous neurons. Classical feedforward models predict that activity will either fade away or become too synchronized unless synaptic strengths fall within a narrow stable regime. Recent studies have shown that biological neural networks can transmit firing-rate information reliably. How do neural circuits find and maintain the operating regime needed for stable rate-code transmission? We propose a simple local mechanism, relational homeostatic scaling, in which each neuron adjusts the strength of its incoming excitatory synapses by comparing a recent trace of its input activity with a trace of its own activity. Unlike conventional homeostatic scaling, the goal is not to force every neuron toward a fixed firing rate. The goal is to preserve the relationship between input drive and output activity. We show that this rule allows layered spiking networks to self-organize into the stable transmission regime, remains robust to neuronal heterogeneity, and works alongside Hebbian spike-timing plasticity. These results suggest that homeostatic plasticity may help establish the circuit conditions under which experimentally observed rate-code transmission can occur.