The Role of Problem Classification in Online Meta-cognition

Agents operating in open environments must be able to adapt their processing to available resources, deadlines, their goal criteria, and their current problem solving contexts. This paper describes the role of meta-cognition in this process; in particular, we define a meta-cognition framework that uses Naive Bayesian classification of the agent´s current context in order to represent the meta-level control problem as a Markov decision process.