Noise Enhances Multisensor Olfactory Predictions

We address the problem of inferring the location of a target that releases odor in the presence of turbulence. Input in the form of binary odor detection is provided by many sensors placed within the turbulent plume.

Drawing inspiration from biology, we ask whether accuracy of the predictions is affected by positional noise. Using Bayesian inference we find that, surprisingly, noise improves prediction accuracy and that an optimal noise exists that grows with distance from the target. Through an asymptotic theory, we show that noise helps by enabling the agent to leverage additional information resulting from geometry of the odor plume. As predicted by the theory, positional noise is beneficial only in anisotropic plumes, both for uncorrelated measures and for turbulent plumes featuring strong spatiotemporal correlations. While precise error tuning is not possible in practice, the asymptotics suggests an adaptive error tuning that approximates well the optimal error.

We extend the analysis to show that different kinds of positional noise as well as sensory noise also help making more accurate predictions, owing to their ability to break the spatiotemporal correlations of the turbulent plume.