Deep Reinforcement Learning with Feedback-based Exploration