Learning by analogical replay in prodigy: First results

Abstract

Robust reasoning requires learning from problem solving episodes. Past experience must be compiled to provide adaptation to new contingencies and intelligent modification of solutions to past problems. This paper presents a comprehensive computational model of analogical reasoning that transitions smoothly between case replay, case adaptation, and general problem solving, exploiting and modifying past experience when available and resorting to general problem-solving methods when required. Learning occurs by accumulation and reuse of cases (problem solving episodes), especially in situations that required extensive problem solving, and by tuning the indexing structure of the memory model to retrieve progressively more appropriate cases. The derivational replay mechanism is briefly discussed, and extensive results of the first full implementation of the automatic generation of cases and the replay mechanism are presented. These results show up to a 20-fold performance improvement in a simple transportation domain for structurally-similar problems, and smaller improvements when a rudimentary similarity metric is used for problems that share partial structure in a process-job planning domain and in an extended version of the STRIPS robot domain.