Deep Reinforcement Learning: theory, intuition, code
summary
Claims concrete implementations of DQN PG and A3C less than 200 lines of python using Keras and shares a code link; demonstrates Atari gameplay with a convolutional net (~3 million parameters) that learns after about a day of training and a browser-navigation benchmark that solves a calendar task in roughly an hour with a tiny actor‑critic net (990k params). Emphasizes the exact losses used (policy value entropy), parallel envs for A3C, and practical monitoring tips.
AI-written summary · about
chapters (8)
- 0:00Introduction to deep reinforcement learning
- 2:15Why RL matters
- 6:18RL practical demo
- 8:30Neural network approximations
- 12:58Actor-critic method
- 19:20A3C implementation
- 22:33RL for web navigation
- 27:28Results and future
related talks
01h 17mTraining intelligent game agents using deep reinforcement learning
26mAnkur Narang; Senior Vice President – Technology & Decision Sciences at Yatra at Cypher2017- Part1/3
56mA Tutorial on Reinforcement Learning II
44mReinforcement learning: A gentle introduction
30mReinforcement Learning - ChatGPT, Playing Games & More
33mBreakthroughs and the Future of (Deep) Reinforcement Learning
49mML: Teaching machine to learn from its own mistakes (...)
01h 03mIEEE CoG 2019 - Industry Day - David Silver (DeepMind)