Convergency of learning process

This paper presents a learning process analysis on stability of learning in light of iterated belief revision. We view a learning process as a sequential belief change procedure. A learning policy is sought to guarantee every learning process leads to a complete knowledge about the world if the newly accepted information is the true fact on the world. The policy allows an agent to abandon the knowledge it has learned but requires a relatively moderate attitude to new information. It is shown that if new information is not always accepted in an extremely skeptical attitude and the changes of belief degrees follow the criterion of minimal change, any learning process for learning truth will converge to a complete knowledge state.