A SELF-TRAINING SYSTEM THAT LEARNS THROUGH EXPERIMENTATION
The paper introduces an adaptive system that, inspired by the diversity of human cognitive development processes, uses different kinds of machine learning to develop its expertise. The system combines a supervised learning recognition engine with an autonomous learning agent. Based on the system’s initial knowledge novel training sets are produced through „experimental learning“. Thus, the user does not have to spend time on the generation of training data. Both the conceptual basis and an exemplary implementation (recognition and substitution of hand-drawn geometrical shapes) are presented. The dependency of the system’s learning success on the knowledge initially provided and the way it processes this knowledge is tested and documented.