Is a Good Representation Sufficient for Sample Efficient Reinforcement Learning - Sham Kakade