Continuous attractor neural network model of multisensory integration

Abstract

In real bio-systems, animals can combine multiple independent sources of information (sensory cues) to reduce uncertainty and improve perceptual performance. Despite intense recent interest in cue integration, the underlying neural mechanisms remains unclear. Continuous attractor neural network (CANN) can be interpreted as an efficient framework for implementing population coding and decoding. In this work, we show that CANN can account for many empirical principles in multisensory integration. Viewed from single neuron behaviors, CANN model can account for the principle of inverse effectiveness and the spatial principle. From the perspective of the activities of population of neurons, CANN can account for the mathematical rule by which multisensory neurons combine their inputs with respect to different cue reliabilities.