A Computational Neural Theory of Multisensory Spatial Representations

Abstract

Multisensory integration refers to the process by which signals coming from distinct sensory systems, but potentially related to the same physical object, are combined. This chapter begins by reviewing the standard neural theory of multisensory integration. Although this theory accounts for some of the anatomical and neurophysiological data, it fails to explain the existence of a type of receptive field commonly found in multisensory areas, known as a partially shifting receptive field. Moreover, it does not fully resolve the recoding problems and neglects statistical issues altogether. The chapter proposes an alternative view based on a neutral architecture using a combination of basis functions and attractor dynamics. The basis function units are used to solve the recording problem, while the attractor dynamics are used for optimal statistical inferences. This architecture accounts for the partially shifting receptive fields, which emerge naturally as a result of the network connectivity and dynamics. Finally, it is shown how this theory may account for the behaviour of human participants when reaching toward multisensory targets.