Zaslat SMS: Adaptive representations for reinforcement learning