Patching Approximate Solutions in Reinforcement Learning

This paper introduces an approach to improving an approximate solution in reinforcement learning by augmenting it with a small overriding patch. Many approximate solutions are smaller and easier to produce than a flat solution, but the best solution within the constraints of the approximation may fall well short of global optimality. We present a technique for efficiently learning a small patch to reduce this gap. Empirical evaluation demonstrates the effectiveness of patching, producing combined solutions that are much closer to global optimality.