Sensory fusion: a neurocomputational approach

The merging of senses is one of the most impressive capacities of the brain. Understanding how this ability can be realized by biologically inspired neural circuits is of the greatest value not only in cognitive neuroscience, but also in neuroclinics (for the design of innovative rehabilitation procedures and neuroprostheses) and in many engineering problems. The present study describes two neural networks inspired by brain functioning, to simulate the main characteristics of audio- visual multisensory integration. The first, motivated by the present knowledge on the Superior Colliculus (a midbrain structure that drives reflexive movements) is based on a feedforward schema, and can simulate many properties of audio-visual integration. The second, inspired by perception in the cerebral cortex, also includes a feedback between the auditory and visual activities, and realizes a Bayesian inference, able to estimate stimuli positions and their causal structure in a near-optimal way.