Modern Reinforcement Learning: Actor-Critic Methods