Reinforcement learning: A gentle introduction
summary
Starts from Atari, AlphaGo Zero, Dota2 and StarCraft as motivations then explains reinforcement learning basics: state action reward interface, Markov decision processes, return G with discount factor gamma, and the policy/value/Q distinction. Covers Monte Carlo policy evaluation (sample episodes, average returns), temporal-difference / TD(0), Q-learning and epsilon‑greedy exploration, policy gradients and actor‑critic, deep Q‑networks using neural nets as function approximators, OpenAI Gym and Python/Numpy example code (not building AlphaGo Zero).
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:31A confession
- 2:11Motivation
- 5:20Core idea
- 6:45Interface
- 7:40Modeling
- 8:33Questions
- 10:10The model
- 14:13How to learn
- 15:59Maximizing rewards
- 19:39Optimal policies
- 20:30Generalized policy iteration
- 22:13Monte Carlo
- 27:06Explorer Exploit Dilemma
- 29:28Temporal Difference Learning
- 32:05Deep Neural Networks
- 34:36Policy gradients
- 37:14Where to start
- 41:40QA
related talks
43mIntroduction to (deep) reinforcement learning
49mReinforcement Learning: Demystifying the hype
30mReinforcement Learning - ChatGPT, Playing Games & More
49mML: Teaching machine to learn from its own mistakes (...)
28mDeep Reinforcement Learning: theory, intuition, code
01h 03mIEEE CoG 2019 - Industry Day - David Silver (DeepMind)
56mAn Introduction to Reinforcement Learning - Snake Your Way Out of a Paper Bag
33mBreakthroughs and the Future of (Deep) Reinforcement Learning