Decomposition of Reinforcement Learning for Admission Control of Self-Similar Call Arrival Processes

This paper presents predictive gain scheduling, a technique for simplifying reinforcement learning problems by decomposition. Link admission control of self-similar call traffic is used to demonstrate the technique. The control problem is decomposed into on-line prediction of near-future call arrival rates, and precomputation of policies for Poisson call arrival processes. At decision time, the predictions are used to select among the policies. Simulations show that this technique results in significantly faster learning without any performance loss, compared to a reinforcement learning controller that does not decompose the problem.