Learning Representations For Efficient Exploration And Goal-Conditioned Reinforcement Learning