Cheyne Weis, University of Chicago

Non-Reciprocity Induced Avalanching in Sparse Networks of Spiking Neurons

Networks of sparsely connected excitatory and inhibitory spiking neurons display a range of dynamical phases, from slow, irregular firing to globally synchronized activity. Neural systems have asymmetric nonreciprocal interactions between excitatory and inhibitory neurons, distinct from equilibrium models in statistical physics. To explore these dynamic phases, we approximate spiking neurons using a theta model, where nonreciprocity manifests as an angular drive that manifests the nonequilibrium properties of the model. We demonstrate that at large systems sizes the nonreciprocity between neurons and thermal noise renormalize to dominate the system beyond a critical nonequilibrium length scale, thereby eliminating inactive regimes present at finite sizes. Our findings are consistent across simpler maximum entropy models, which we use to investigate inference near the boundary between inactive and active states. We observe that the quality of inference peaks near this boundary, implying that maximal information transfer could be occurring at the crossover between phases. This suggests that neurons operating near an avalanching boundary could optimize information transfer, enhancing the ability of downstream neurons to decode stimuli from upstream neurons.