Ágnes Vathy-Fogarassy, PhD; Tibor Dulai, PhD
Prerequirements:
-
Topics covered:
Students will learn about the following topics, including their individual training plans and interests:
T1. The basic concepts of reinforcement learning:
- the concept and characteristics of reinforcement learning and related concepts (reward, environment, state, history)
- the main components of reinforcement learning (policy, value function, model)
- problem areas within reinforcement learning (learning - planning, exploration - exploitation, prediction - control, on-policy learning - off-policy learning)
T2. Markov processes and model-based solution methods:
- Markov Reward Processes and their solutions
- Markov Decision Processes and their solutions
- comparing the Bellman Expectation Equation and the Bellman Optimality Equation in terms of their role and form
- planning by dynamic programming (policy evaluation, policy iteration, and value iteration)
T3. Model-free reinforcement learning:
- model-free prediction (Monte Carlo Learning, Temporal-Difference Learning, and TD(λ))
- model-free control (on-policy methods: Monte Carlo Learning, on-policy Temporal-Difference Learning, and Sarsa(λ); off-policy Learning: importance sampling and Q-learning)
Literature:
1. Richard S. Sutton, Andrew G. Barto: Reinforcement learning: An introduction, The MIT Press, 2015.
2. David Silver: RL Course by David Silver, DeepMind, 2015.