Representational similarity modulates neural and behavioral signatures of novelty

Abstract

Novelty signals in the brain modulate learning and drive exploratory behaviors in humans and animals. Inherently, whether a stimulus is novel or not depends on existing representations in the brain, yet it remains elusive how stimulus representations influence novelty computation. In particular, existing models of novelty computation fail to account for the effects of stimulus similarities that are abundant in naturalistic environments and tasks. Here, we present a unifying, biologically plausible model that captures how stimulus similarities modulate novelty signals in the brain and influence novelty-driven learning and exploration. By applying our model to two publicly available data sets, we quantify and explain (i) how generalization across similar visual stimuli affects novelty responses in the mouse visual cortex, and (ii) how generalization across nearby locations impacts mouse exploration in an unfamiliar environment. Our model unifies and explains distinct neural and behavioral signatures of novelty, and enables theory-driven experiment design to investigate the neural mechanisms of novelty computation.