Approximate strategies for learning trajectories of autonomous learning agents

Describes a new approach to the approximate modeling of the learning dynamics of autonomous learning agents for performance improvement in supervised learning. The extracted approximate model can be used to generate target trajectories from the current performance state to the final performance goal in order to “lead” the learning agent through the dynamic range of the learning process. The interaction between the supervisor module and the agent can be modeled as an incentive game. Ideas introduced for the single-agent case can further be extended to include multi-agents to address the coordination problem in modular learning structures