Learning to Detour & Schema-Based Learning

摘要: In this paper we extend the analysis of learning to detour (Corbacho and Arbib, 1995) by proposing a generalized framework Schema-based (SBL) which incorporates general principles adaptive organization e.g., bootstrap coherence maximization principles. A schema is an evolutionarily or experience-based constructed recurrent pattern interaction expectation (perceptual, motor, reactive, predictive schemas) with environment, measure congruence between result environment expectations agent has for that interaction. SBL attempts provide formal independent particularities implementation, tius allowing design wide variety animats. allows growth increasingly complex patterns from initially restricted stock schemas. also efficient confining statistical estimation narrow credit assignment space, so system learns in right ballpark minimum “scene” statistics while maintaining high degree adaptability.