Learning through explaining observed inconsistencies

摘要

Perpetual learning is an essential capability for long-lived cognitive agents (natural or artificial) to survive in dynamic and changing environments. Previous work on inconsistency-induced learning, iLearning, has proposed a general framework for perpetual learning agents where learning amounts to finding ways to circumvent inconsistencies. This paper continues the ongoing research of iLearning by defining observed inconsistencies and describing a learning algorithm that reconciles observed inconsistencies through finding some viable explanation. We compare our work with related work on life-long learning, learning through resolving anomalies, and truth finding problem.