Local homeostatic scaling supports stable rate propagation under noise and heterogeneity

Abstract The brain maintains stable information transmission despite substantial heterogeneity in neuronal properties and noise. Signals must travel across multiple regions with high fidelity without interruption or distortion. Previous studies achieve reliable firing-rate propagation in feedforward circuits only with finely tuned synaptic weights; otherwise, firing rate activity either decays or explodes. How neural circuits self-organize to achieve stable rate propagation without external fine-tuning remains unclear. Here, we introduce a simple, local homeostatic synaptic scaling rule in which each neuron rescales its excitatory inputs based on the mismatch between mean excitatory presynaptic and postsynaptic firing rate. We show both analytically and numerically that this local learning rule enables noisy feedforward networks to spontaneously organize stable rate propagation across layers when driven by external inputs, without the need for global supervision or precise weight initialization. This rule compensates for intrinsic neuronal properties by adjusting synaptic strengths to maintain consistent firing rate transmission, which allows heterogeneous feedforward networks to self-organize to support stable rate propagation. Finally, we show that when combined with homeostatic inhibitory synaptic plasticity, the rule stabilizes attractor firing rate in recurrent networks, preserving an asynchronous-irregular activity under perturbations. Our findings reveal how local plasticity rules support robust neural circuits capable of signal propagation without external fine-tuning, providing mechanistic insights into neurobiological self-organization and principles for designing resilient spiking networks. Author Summary The brain is made of diverse and noisy parts, yet information still needs to move reliably from one area to the next. Many studies explain this by assuming that synaptic strengths are set with high precision. Such fine tuning is unlikely in real circuits. We used models and theory to ask whether simple, local rules could remove this need for precision. We found that if each neuron rescales the strengths of its incoming excitatory synapses according to ongoing activity, feedforward pathways teach themselves to pass firing rates from layer to layer. This self-organization is robust: it continues to work even when neurons are noisy and differ from one another. When the same idea is coordinated with a plastic change at inhibitory synapses, recurrent networks keep stable patterns of activity and remain in a healthy, irregular regime even after disturbances. Our study suggests that coordinated homeostatic plasticity can wire, protect, and stabilize the brain’s information highways without global control.