Annotation

The course provides a comprehensive introduction to deep reinforcement learning, a combination of reinforcement learning and deep neural networks that demonstrated super-human capabilities in diverse domains, including complex games like Go and chess, chip design, automated discovery of algorithms and architectures, and advancing robotics and LLMs. The course focuses both on theory and practical implementations in PyTorch through weekly assignments solved individually or in small teams.

Location

Malostranské náměstí 25 , 118 00, Praha

Results of learning

Knowledge: The Graduate describes and explains the basic concepts of reinforcement learning (MDP, policy, value function), characterizes the main approaches (Monte-Carlo/temporal difference, on-policy/off-policy, tabular/function approximation, value-based/policy-gradient methods, model-free/model-based), describes the algorithms and their corresponding architectures (DQN, Rainbow, REINFORCE, A3C, PAAC, DDPG, SAC, PPO), and explains planning-based approaches (MCTS, AlphaZero, MuZero, Dreamer).

Skills: The Graduate implements the aforementioned basic architectures in the PyTorch framework and is able to train them effectively. The Graduate understands a scientific paper from the field of deep reinforcement learning.

Competence: The Graduate proposes a method to solve a new (previously unknown to them) problem from the field of reinforcement learning, and implements and evaluates it themselves.

Contact for applicants

magdalena.kokesova@matfyz.cuni.cz