Sensory noise predicts divisive reshaping of receptive fields

Matthew Chalk, Paul Masset, Sophie Denève, Boris S. Gutkin

Abstract: Author summary Perception involves using incoming sensory signals to infer which objects or features are present in the surroundings. To do this, sensory systems must perform two basic operations: (i) combination of noisy sensory cues, and (ii) competition between different percepts. Here we show that the optimal form of competition depends on how sensory signals are corrupted by noise. Moreover, for the type of noise commonly observed in sensory systems, whose variance scales with the signal amplitude, competition should occur between different sensory cues before they are combined. Implemented neurally, this results in a highly flexible representation, in which neural receptive fields change dynamically depending on the stimulus context. Further we show that competition should take the form of divisive inhibition from the surround, accounting for why divisive normalisation, gain control and contrast dependent temporal dynamics appear so ubiquitous in sensory areas.