QuizMe

pt-drl

מנהלים:

Rahaf Sbeh

		Behavior Cloning (BC): Directly learns from expert trajectories (supervised learning).
		DAgger (Dataset Aggregation): Interactively improves policy by incorporating expert corrections.
		Inverse RL (IRL): Learns the reward function from expert demonstrations.

		Epsilon-Greedy: Explores randomly some percentage of the time.
		Upper Confidence Bound (UCB): Explores actions with high uncertainty.
		Thompson Sampling: Uses Bayesian inference to balance exploration/exploitation.

		Combines MCTS with deep neural networks for playing Go.
		Uses policy and value networks for better search efficiency.