Using Bayes' rule to model multisensory enhancement in the superior colliculus.

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• The deep layers of the superior colliculus (SC) integrate multisensory inputs to initiate orienting responses.
• Multisensory enhancement, a phenomenon in the deep SC, augments neural responses by combining inputs from different modalities.
• This enhancement is behaviorally linked to increased likelihood of orienting toward weak stimuli when multiple sensory cues are present.

Purpose of the Study:

• To investigate the computational principles underlying multisensory integration in the deep SC.
• To determine if neurophysiological findings in the deep SC support a probabilistic model of target detection.
• To explain the phenomenon of inverse effectiveness in multisensory neurons.

Main Methods:

• Modeling multimodal sensory inputs to the deep SC as random variables.
• Applying Bayes' rule to formulate the computation function for multisensory neurons.
• Analyzing the relationship between unimodal response strength and the degree of multisensory enhancement.

Main Results:

• Deep SC neurons compute the probability of a target's presence using integrated sensory inputs.
• Multisensory neurons preferentially combine unimodal inputs that are individually more uncertain.
• Inverse effectiveness arises because integrating multisensory inputs yields a larger increase in target probability when unimodal responses are weaker.

Conclusions:

• Deep SC neurons act as probabilistic inference engines, integrating multisensory information to estimate target presence.
• The properties of multisensory enhancement and inverse effectiveness are consistent with a Bayesian computation framework.
• This framework provides a unified explanation for how the deep SC utilizes multisensory information for adaptive behavioral responses.