Optimal noise level for coding with tightly balanced networks of spiking neurons in the presence of transmission delays

Jonathan Timcheck, Jonathan Kadmon, Kwabena Boahen, Surya Ganguli

Abstract: Author summary Remarkably, the brain can perform precise computations, despite being composed of noisy neurons with slow, unreliable synaptic connectivity. To understand how this is possible, we can imagine the classic strategy where neurons are grouped into weakly-coupled subpopulations, creating redundancy to achieve high precision. But interestingly, recent work proposed a tight-balance neural network that instead uses fast, strong connectivity between neurons to achieve much higher precision with the same number of neurons. This efficiency is attractive, but notably, signals take time to propagate in the brain. Such propagation delays alone can lead to pathological synchronization. Intriguingly, while noise commonly degrades the performance of a computational system, it has been observed in simulations that noise can help mitigate synchronization and in fact rescue performance in tight-balance networks. In this work, we develop a theory that quantifies the simultaneous effects of delays and noise in tight-balance networks, and allows us to compute the optimal noise level as a function of delay, yielding conceptual insights into how noise can counteract delay induced synchronization to preserve precise computation in efficient neural networks.