| Abstract: |
Direct Policy Search (DPS) is a method for machine learning, where the policy is represented by some parameters and they are optimised by evaluating the policy directly. We introduce Exemplar Based Policy (EBP) Optimisation as a framework of DPS. An EBP is a policy composed of the set of exemplars (parameters to be optimised) and a case-based action selector. An actual implementation with GA optimiser, Genetic Policy Search, is introduced and its performances for two problem domains are shown. |