Habituation to strongly fluctuating odor backgrounds

Animals rely on their sense of smell to survive, but important olfactory cues are mixed with odor backgrounds that fluctuate strongly, due to turbulence. It is unclear how the olfactory system can habituate to such stochastic backgrounds to better detect new, relevant odors. While studies of adaptation in other sensory systems have focused on predictive filtering, here we show that, in the high-dimensional olfactory space, background manifold learning offers an improved theoretical performance for habituation and new odor recognition. We then propose models of manifold learning in the early olfactory circuit, based on a layer of interneurons learning input projections via either the IBCM (Intrator and Cooper, 1992) model of synaptic plasticity, or a biological online PCA rule (Minden et al., 2018). We then examine the performance of these models for habituation to turbulent backgrounds and find that, after introducing proper scaling parameters, they both perform significantly better than simple predictive mean filtering. We analyze in particular the IBCM network, to find that each interneuron selectively inhibits one background odor. IBCM interneurons thus encode biologically relevant information about individual odors, which could be leveraged by more elaborate circuits.