Forgetting and aging of knowledge in concept formation

One of the problems solved by machine learning based techniques is symbolic data analysis and concept formation. We report on the program FAVORIT, which achieves a performance improvement over its predecessors (such as UNIMEM) by means of a simple mechanism mimicking the shortcomings of human learning: aging of knowledge and forgetting. When applied to large and noisy data sets, these characteristics enable efficient restructuring and pruning of the internal knowledge structures. The paper contains a brief description of the program, together with the rationale behind its philosophy, as well as a simple case study.