On Natural Statistics in Multisensory Integration

Multisensory neurons in the deep layers of the superior colliculus (DSC) integrate different sensory modalities, responding more strongly to spatially congruent inputs from distinct modalities than to input from a single modality (cross-modal enhancement, or CME) [1]. Multiple stimuli presented within a unimodal receptive field, however, evoke a weaker response than a single stimulus presented in the same receptive field (modality-specific suppression, or MSS) [2]. Although CME and MSS have been widely studied experimentally, the mechanisms underlying these responses remain unclear. As sensory neurons, multimodal DSC neurons may employ the same adaptive processes as unimodal neurons earlier in sensory pathways by adapting to the statistical properties of their inputs (cf. [3]) through processes such as gain control and firing threshold adjustment. Here we report, from a theoretical perspective, on the role that input statistics may play in multisensory neurons.