Introduction to Reinforcement Learning
- Erogato da: UCL / DeepMind
- Docente: David Silver
- Livello: intermediate
- Formato: video-lectures
- Durata: 10 Lectures
- Categoria: reinforcement-learning
- Lingua: English
- Prezzo: Gratuito
- Tag: reinforcement-learning, mdp, dynamic-programming, policy-gradient, value-function, exploration
Questa classica serie di lezioni tenuta allo University College London da David Silver, Principal Scientist di DeepMind e co-creatore di AlphaGo e AlphaZero, introduce i principali metodi e tecniche dell'apprendimento per rinforzo. Il corso di 10 lezioni tratta: introduzione all'RL, processi decisionali di Markov, programmazione dinamica, predizione model-free, controllo model-free, approssimazione della funzione di valore, metodi policy gradient, integrazione di apprendimento e pianificazione, esplorazione e sfruttamento, e un caso di studio sull'RL nei giochi classici.
Perché seguirlo
Taught by David Silver — the lead researcher behind AlphaGo and AlphaZero — this lecture series is the most widely recommended introduction to reinforcement learning. Silver's ability to explain complex concepts clearly, combined with his unique perspective from building systems that defeated world champions at Go, makes this an unparalleled learning resource. It has introduced thousands of practitioners to RL.
Cosa imparerai
- The reinforcement learning problem formulation and Markov Decision Processes
- Dynamic programming methods for solving known MDPs
- Model-free prediction: Monte Carlo and temporal-difference learning
- Model-free control: SARSA, Q-learning
- Value function approximation with neural networks
- Policy gradient methods and the REINFORCE algorithm
Programma
Ten self-contained lectures progressing from fundamentals to advanced topics. Starts with the RL problem formulation and MDPs, builds through dynamic programming and model-free methods (both prediction and control), then covers function approximation, policy gradients, planning integration, and exploration. Concludes with a fascinating case study applying RL to classic games including the AlphaGo breakthrough.
Argomenti principali
- Introduction to reinforcement learning
- Markov Decision Processes (MDPs)
- Dynamic programming: policy iteration, value iteration
- Model-free prediction: MC, TD(0), TD(λ)
- Model-free control: on-policy and off-policy
- Value function approximation
- Policy gradient methods
- Integrating learning and planning (Dyna, MCTS)
- Exploration vs. exploitation
- Case study: RL in classic games (Atari, Go)
Prerequisiti
- Basic probability and statistics
- Linear algebra fundamentals
- Some programming experience (for understanding pseudocode)
A chi si rivolge
- Anyone new to reinforcement learning seeking a strong theoretical foundation
- Students preparing for more advanced RL courses (like CS 285)
- Practitioners wanting to understand the fundamentals before applying RL libraries
- Researchers from other fields looking to add RL to their toolkit
Link
- Sito web: https://www.davidsilver.uk/teaching/