Learning through overcoming temporal inconsistencies

Perpetual learning is an indispensable capability for long-lived intelligent agents (natural or artificial) to adapt to dynamic and changing environments. In our previous work on inconsistency-induced learning, i2Learning, we have proposed a general framework and several inconsistency-specific learning algorithms for perpetual learning agents that consistently and continuously improve their performance at tasks over time through overcoming inconsistencies. This paper reports the latest results of the i2Learning research on treating temporal inconsistencies as learning stimuli and defining a learning algorithm that improves an agent's performance through refining its problem-solving knowledge as a result of overcoming temporal inconsistencies the agent encounters. We also compare our approach with related work.