UC Berkeley · Sergey Levine · Fall 2023 recordings

Deep RL, from imitation to open problems.

All 22 recorded lectures of CS 285 (plus four guest lectures) rebuilt as structured lessons — Levine's own arc, reorganized for reading, with the derivations typeset, key remarks pulled out, multi-part videos in one player, and exercises to check yourself. The five homeworks get companion pages that point you at the real work. Watch, read, work, tick it off.

Start with Lecture 01 Official course site ↗ Search the course
Part 1

Foundations: Imitation to Policy Gradients

  1. 01 Introduction
  2. 02 Imitation Learning
  3. 04 Introduction to RL
  4. 05 Policy Gradients
  5. ★ HW1 · Imitation Learning
Part 2

Value-Based Methods

  1. 06 Actor-Critic Algorithms
  2. 07 Value Function Methods
  3. 08 Deep RL with Q-Functions
  4. 09 Advanced Policy Gradients
  5. ★ HW2 · Policy Gradients
  6. ★ HW3 · Q-Learning & Actor-Critic
Part 3

Model-Based RL and Exploration

  1. 10 Optimal Control and Planning
  2. 11 Model-Based RL
  3. 12 Model-Based RL with Policies
  4. 13 Exploration I
  5. 14 Exploration II
Part 4

Offline RL and Theory

  1. 15 Offline RL I
  2. 16 Offline RL II
  3. 17 RL Theory
  4. ★ HW4 · LLM RL
  5. ★ HW5 · Offline RL
Part 5

RL as Inference

  1. 18 Variational Inference
  2. 19 Control as Inference
  3. 20 Inverse Reinforcement Learning
Part 6

Frontiers

  1. 21 RL with Sequence Models & LLMs
  2. 22 Transfer Learning & Meta-Learning
  3. 23 Challenges & Open Problems
Part 7

Guest Lectures

  1. ★ Eric Mitchell · RLHF
  2. ★ Andrea Zanette · Statistical Foundations of RL
  3. ★ Aviral Kumar
  4. ★ Dorsa Sadigh

Personal study companion for UC Berkeley CS 285 (Prof. Sergey Levine). Lectures, slides, and homeworks © UC Berkeley and the CS 285 staff. Private, non-commercial study use; not affiliated with Berkeley. Recordings are the Fall 2023 offering; slides link to the current revision. Lecture 3 was not recorded.