Hierarchical Combination of Bayesian Models and Representations

Abstract

Ever since seminal work by Marr [11] and Fodor [10] up until more recent accounts such as given by Ballard [8] and many others on computational theories of perception and cognition, the link between the functional organization of perceptual sites in the brain and the underlying computational processes has led to the belief that modularity plays a major role in making these processes tractable. Modularity, in this sense, means that the flow of computation can be broken down into simpler processes. As a matter of fact, although the interconnections between these sites have increasingly been found to be much more intricate than Marr believed (including feedback and lateral links), the notion that the brain is organised in a modular fashion has been supported by countless findings in Neuroscience research, and is currently undisputed.