A neural architecture for autonomous learning

Psychology and neurobiology nowadays provide a large amount of precise info rmation on visual system function. This information can be used in the desi gn of autonomous systems capable of learning and recognising objects and pl aces important for survival in complex unknown (real or virtual) environmen ts. Our work is based on the principles that perception is fundamentally a dynamic process in constant interaction with movement; and that learning ca n be made simpler if the systems are not required to learn the invariants o f their environment (e.g. preservation of neighbour topological relations, or connectivity of the space). The techniques that contribute to devising t hese adaptive systems in continuous interaction with their environment coul d significantly influence our approach to programming and the man-machine i nterface.