Abstract
In our previous work, we described a framework called i 2 Learning for a perpetual learning agent to be engaged in continuous learning to incrementally improve its problem solving performance over time. i 2 Learning offers an overarching framework that can accommodate various inconsistency-specific learning strategies. In this paper, we report our new results on how learning can be carried out through overcoming inheritance inconsistencies the agent encounters during its problem-solving episodes. Each learning episode causes the agent's knowledge to be refined or augmented so as to overcome the encountered inheritance inconsistency. This will in turn improve the agent's performance at tasks incrementally. The work in this paper is an integral part of the overall effort toward fully developing the i 2 Learning framework.