The Neural Aesthetic @ ITP-NYU, Fall 2018
Lecture 10: Reinforcement Learning & Natural Language Processing [11/27/2018]
[Slides]
- Introduction to natural language processing (NLP) (5:40)
- Why is NLP hard? (10:02)
- Word embeddings (12:02)
- Properties of word vectors (21:10)
- Tutorial: universal sentence encoder (28:24)
- Applications of sentence embeddings (37:21)
- Machine translation (43:53)
- Tutorial: Wikipedia latent semantic analysis (LSA) (50:47)
- spaCy tutorial (1:00:24)
- Introduction to reinforcement learning (RL) (1:06:23)
- The RL setup (1:09:07)
- Examples and challenges of RL problems (1:12:03)
- Deep Q-Networks for beating Atari games (1:16:22)
- Applications to robotics and humanoid simulation (1:26:24)
- Monte Carlo tree search (MCTS) (1:31:06)
- Tic-tac-toe MCTS (1:33:57)
- Introduction to Go and AlphaGo (1:42:18)
- How AlphaGo improves MCTS (1:50:18)
- AlphaGo vs. Lee Sedol (1:55:19)
- AlphaGo Zero and discarding training data (1:58:40)
- AlphaZero generalized (2:05:03)
- AlphaZero plays chess and crushes Stockfish (2:09:55)
- Curiosity-driven RL exploration (2:16:26)
- Practical resources for reinforcement learning (2:18:01)