#imitation-learning

Question 1: Single choice

To build an Othello player, you train a policy network on a dataset of expert games: for each board the network outputs log-probabilities over the board positions and is trained to maximize the log-probability of the move the expert made. What kind of learning is this?

Behavioural cloning: supervised learning of the expert's moves, without any rewards1
Q-learning: the network learns the value of each move from the final game outcome2
Policy-gradient RL: moves are reinforced in proportion to the return of each game3
Unsupervised learning: the network clusters board positions without any targets4

Timed test