Heterogeneous Time Constants Improve Stability in Equilibrium Propagation

TL;DR

This paper introduces heterogeneous time constants in equilibrium propagation, inspired by biological neurons, to improve training stability and robustness without sacrificing task performance.

Contribution

It proposes a novel method of assigning neuron-specific time constants in equilibrium propagation, enhancing biological plausibility and training stability.

Findings

Heterogeneous time constants improve training stability.

The method maintains competitive task performance.

Increases biological realism and robustness.

Abstract

Equilibrium propagation (EP) is a biologically plausible alternative to backpropagation for training neural networks. However, existing EP models use a uniform scalar time step dt, which corresponds biologically to a membrane time constant that is heterogeneous across neurons. Here, we introduce heterogeneous time steps (HTS) for EP by assigning neuron-specific time constants drawn from biologically motivated distributions. We show that HTS improves training stability while maintaining competitive task performance. These results suggest that incorporating heterogeneous temporal dynamics enhances both the biological realism and robustness of equilibrium propagation.