AlphaGo — what made it different from prior Go engines?
mediumAnswer
- Silver et al. 2016.
- (1) Deep NN combining value and policy (previously hand-crafted).
- (2) Trained via supervised learning from human games → reinforcement learning via self-play.
- (3) MCTS guided by NN policy prior + value estimate.
- (4) Beat top human Lee Sedol 4-1 in 2016.
- Predecessors relied on hand-crafted features + Monte Carlo rollouts.
- AlphaZero later dropped the human-game pretraining, learned purely from self-play.
Check yourself — multiple choice
- Random
- Deep NN combining policy + value + SL from human games → RL self-play + MCTS guided by NN prior; beat Lee Sedol 4-1 2016; AlphaZero later dropped SL for pure self-play
- Same as DQN
- Not real
AlphaGo: policy + value NN + SL + RL self-play + MCTS; beat Lee Sedol 2016.
#applications
Practise Reinforcement Learning
214 interview questions in this topic.