Unsupervised learning of eye-hand-coordination.
Andreas Birk, Wolfgang J. Paul · 1999
. Unsupervised learning of eye-hand-coordination is an interesting problem for two very different reasons. First, it is an important step in the human cognitive development. Second, when applied to a real-world set-up as we do here, it has an application potential. Demonstrating its potential on this concrete task, we present a novel approach to unsupervised learning, the socalled stimulus-response-learning or short SRL. It features an on-line evolution of simple reactive rules stored in a dynamic directed graph, such that both reactive behavior and a word-model are learned in parallel. The graph is somewhere similar to belief-networks as it represents potential consecutive activation of rules, hence allowing a simple inference of future states of the environment in dependence of possible actions of the system. But the unsupervised learning of both rules and the graph are not based on a bayesian or any related learning technique, but on a novel type of on-line evolution. Unlike common ...