CS 185/285 at UC Berkeley Deep Reinforcement Learning

What You'll Learn
Click Here

Official Source

  • Fundamentals of Reinforcement Learning (RL) and how agents learn through trial and error.

  • Imitation Learning and Behavioral Cloning to learn from expert demonstrations.

  • Policy Gradient methods for directly optimizing decision-making policies.

  • Actor-Critic algorithms that combine value estimation and policy learning.

  • Value-Based RL techniques, including Q-Learning and Deep Q-Networks (DQN).

  • Advanced RL methods such as SAC, advanced policy gradients, and exploration strategies.

  • Variational Inference and Control as Inference for probabilistic decision-making.

  • Model-Based RL for planning using learned environment models.

  • Offline RL, where agents learn from previously collected datasets.

  • LLM Reinforcement Learning, including training and aligning large language models.

  • Multi-task RL, RL theory, and current challenges and open research problems.

  • Practical implementation using PyTorch, along with hands-on homework assignments and a final project.


This course teaches how intelligent agents learn to make decisions through reinforcement learning. You will study core RL algorithms such as Policy Gradients, Actor-Critic, Q-Learning, Model-Based RL, Offline RL, and LLM RL, while gaining practical experience implementing state-of-the-art deep learning and reinforcement learning systems for real-world AI applications.