Is Poisson-like spiking essential to general-purpose neural information processing?

Authors: Rich Pang

Presentation type: Flash talk at SNUFA 2026 online workshop (4-5 Nov 2026)

Abstract

One of the most pervasive and curious features of mammalian neuronal activity is the Poisson-like temporal irregularity of spike trains. Although observed for decades across diverse brain regions and contexts, whether Poisson-like spiking serves a general functional role or is mostly “just noise” remains unknown. Resolving this issue will be crucial to understanding brain function and guiding the development of bio-inspired spiking networks that compute. A longstanding hurdle, however, is that classic models producing irregular spiking through network interactions, surprisingly, become increasingly macroscopically linear in large networks, impeding their ability to perform nonlinear computations. Here I will present an alternative model in which Poisson-like spiking is in fact indispensable to the network’s function as a possibly general-purpose nonlinear computing system. The individual network element is not a single neuron but a biologically inspired group of differently tuned neurons that compete to spike at each timestep via winner-take-all. Winner-take-all is a powerful computational primitive, but its action in large networks has received little investigation. Via simulations and mean-field theory I demonstrate that these networks produce irregular spiking through chaos while simultaneously supporting flexible, programmable nonlinear dynamics that are powered, rather than hindered, by spiking variability, and which persist in large networks. For example, one can program networks to produce combinatorial fixed point attractors, slow limit cycles, or heterogeneous firing rate patterns with diverse timescales—three dynamical primitives thought to subserve memory-dependent computation in the brain—all sustained by irregular spiking. Ultimately, the macroscopic dynamics are “fluctuation-driven”, or instantiated via changes in variances, rather than means of synaptic inputs, which perpetuate themselves through Poisson-like spiking variability. This demonstrates proof-of-concept of a biologically plausible network where irregular spiking is not “just noise” but an essential component of a flexible fluctuation-driven dynamical system well-posed for general-purpose computation, and which can be trained or fit to data.

Are you the author and do you want to add some related links to this page? If so, check out our guide to adding related links and videos.

Are you the author and do you want to add your embedded video to this page? If so, check out our guide to adding related links and videos.