Learning Representations from Pre-synaptic Glomerular Responses for Odor Classification

ICLR 2026 Conference Submission14337 Authors

Keywords: Olfaction, Smell, Odorants, Ligands, Olfactory Bulb, Glomeruli, Physico-Chemical properties.

TL;DR: We show that representations learned from glomerular calcium imaging at the first olfactory synapse in mice enable accurate odor identity classification.

Abstract: Advances in neural sensing technology now make it possible to observe presynaptic responses in the olfactory bulb with high spatial and temporal resolution. In this paper, we approach olfaction (the sense of smell) from a Data Science and AI perspective, focusing on how odor identity can be decoded from neural representations at the first synaptic stage of the olfactory system. Drawing distinctions to color vision, we argue that smell presents unique measurement challenges, including the complexity of stimuli, the high dimensionality of the sensory apparatus, as well as what constitutes ground truth. In the face of these challenges, we argue for the centrality of odorant-receptor interactions in developing a theory of olfaction. Such a theory is likely to find widespread healthcare applications in disease diagnostics, enhance our understanding of smell, and in the longer-term can help us understand how it relates to other senses and language. As an initial use case, we show that machine learning models can learn meaningful representations from calcium imaging of glomerular activations, enabling accurate odorant classification and revealing that pre-synaptic responses at the first olfactory synapse encode rich, discriminative information about odor identity. Additionally, we release 'oMNIST' — a standardized dataset of glomerular responses for public use—designed to catalyze research in classification, representation learning, cross-animal glomeruli alignment and generalization in olfaction.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 14337