Αποστολή με SMS: Adaptive Representations for Reinforcement Learning