Biological model of vision for an artificial system that learns to perceive its environment

Abstract

A computer algorithm is described which implements models of the biological visual mechanisms of the retina, thalamic lateral geniculate and perigeniculate nuclei, and primary visual cortex. Motion and pattern analyses are performed in parallel and interact in the cortex to construct perceptions. The authors hypothesize that motion reflexes serve as unconditioned pathways for the learning and recall of pattern information. The algorithm demonstrates this conditioning through a learning function approximating heterosynaptic facilitation.<>