Integration of perception and learning in the CAEDUS cognitive architecture

An initial approach to integration of perception and learning subsystems for the CAEDUS cognitive architecture is presented. The perception subsystem is able to adaptively modulate the focus of attention of its feature recognition search based on the set of previously detected features and estimated conditional probabilities. The learning subsystem is able to learn models of observed agent behaviors concurrently and incrementally from samples of agent executed actions. A data interface between these subsystems converts feature sets recognized by perception into training instances for learning. Initial results are given showing improvement of learning due to adaptive perception.