Input regularization mechanisms of the olfactory bulb glomerular layer

Authors: J. Forest, K. R. Mama, R. Moyal, M. Einhorn, A. Borthakur, T. A. Cleland

Presentation type: Poster at SNUFA 2023 online workshop (7-8 Nov 2023)

Abstract

Biological and artificial neural networks require statistically predictable inputs for optimal performance. Constraining the distributions of input activation enables the coordinated computational elements of a functional circuit (i.e., neurons, synapses and network motifs) to operate within their effective response ranges. Peripheral sensory systems, however, are necessarily exposed to relatively unconstrained input variance. Consequently, they must transform and regularize these signal patterns –while preserving their information content– before communicating them to downstream regions. In the olfactory system, this function is governed substantially by glomerular layer circuitry within the olfactory bulb. Previous physiological and computational studies have separately characterized bulbar mechanisms that capture and compress broad-ranging variance, regulate contrast, normalize activity, and statistically regularize input distributions. Here, we unite these mechanisms in a common signal conditioning framework using our spiking neural network modeling framework, Sapicore. Specifically, our signal conditioning layer implements (1) global intensity normalization based on nonspecific lateral projections by superficial short axon cells; (2) non-topographical contrast enhancement based on feedforward inhibition and regulated by neuromodulation; and (3) statistical input regularization based upon heterogeneous duplication within columns, arising from heterogeneity in the properties of co-columnar (sibling) mitral cells. The result is a signal conditioning layer that adapts to broad unregulated patterns of external stimulation and transforms raw sensory input into a statistically reliable format with minimal loss of information. We quantify and analyze the concerted properties of this complex circuit, and show that heterogeneity in neuronal and synaptic properties can be a crucial contributor to the function of natural systems embedded in unregulated environments