Constrained inhibitory plasticity stabilizes competitive STDP networks

Abstract

Competitive spiking neural networks commonly rely on fixed lateral inhibition to establish stable competition between excitatory neurons, although inhibitory synapses are themselves plastic in biological cortical circuits. Whether biologically plausible inhibitory plasticity can replace fixed inhibition while preserving stable competitive learning therefore remains an open question. Using an Euler-based implementation of the Diehl-Cook unsupervised MNIST network [1], we compared fixed inhibition, Vogels-style inhibitory spike-timing-dependent plasticity (iSTDP) [2], and several alternative I→E plasticity rules designed to promote stable competition. Vogels-style iSTDP exhibited only a narrow stability regime, with modest parameter changes leading to runaway population activity and loss of stable competitive dynamics. A slow homeostatic inhibitory learning rule delayed these instabilities but ultimately converged toward inhibitory weight saturation. We therefore evaluated a normalized slow-homeostatic rule that combines local homeostatic updates with a constraint on the total inhibitory input received by each excitatory neuron. In proof-of-concept MNIST experiments, the normalized rule remained stable throughout training, prevented inhibitory weight saturation, and maintained functional classification performance together with firing-rate, inhibitory-conductance, and neuron-utilization statistics comparable to the best stable Vogels-iSTDP regime. These findings suggest that competitive STDP networks may require explicitly constrained inhibitory plasticity rather than unconstrained local inhibitory learning alone, consistent with the view that stable representation learning in spiking neural networks emerges from the coordinated interaction of Hebbian plasticity and homeostatic mechanisms [2–4].

References
[1] Diehl, P. U., & Cook, M. (2015). Unsupervised learning of digit recognition using spike-timing-dependent plasticity. Frontiers in Computational Neuroscience, 9, 99., 10.3389/fncom.2015.00099
[2] Vogels, T. P., Sprekeler, H., Zenke, F., Clopath, C., & Gerstner, W. (2011). Inhibitory plasticity balances excitation and inhibition in sensory pathways and memory networks. Science, 334(6062), 1569-1573., 10.1126/science.1211095
[3] Zenke, F., Agnes, E. J., & Gerstner, W. (2015). Diverse synaptic plasticity mechanisms orchestrated to form and retrieve memories in spiking neural networks. Nature Communications, 6, 6922., 10.1038/ncomms7922
[4] Turrigiano, Gina G. (2012). Homeostatic synaptic plasticity: local and global mechanisms for stabilizing neuronal function. Cold Spring Harb Perspect Biol. 4(1):a005736., 10.1101/cshperspect.a005736