A Computational Investigation of the Evolution of Vision

The computational syntax and terminology proposed by Lamontagne(1987) were used to investigate the problem of the phylogenesis of visual perception. In particular, an attempt was made to specify Lamontagne's principle of adjacency. This principle states that a relevant natural mutation will cause an offshoot to have one or more levels of informational grouping than its parents, where each specific group contains units of information which are adjacent, along one or many continua, to the pivotal unit of that group. A number of assumptions were proposed to constrain the investigation space. Among them was the pairwise-grouping hypothesis, a hypothesis belonging to the set of possible instances of the adjacency principle. This hypothesis states that any formal epistemic entity of level n--that is, n:EE--receives input only from pairs of (n-1):EEs. Given a two amplitude retinal domain, the assumptions were found to predict offspring capable of detecting line segments and some of their orientations. With a multiamplitude retinal domain, however, the pairwise-grouping hypothesis and the general principle of adjacency were found to have difficulty in accounting for smooth contrast detection. Formal neural network solutions were proposed to overcome that and related difficulties.