Reinforcement Learning: An Introduction

Livro digital

Título:
Reinforcement Learning: An Introduction

Autor:
Richard S. Sutton, Andrew G. Barto

Categoria:
Tecnologia > IA

Doador:
Raffaello D. N.

Sinopse:
Reinforcement learning starts with a hard practical problem, how an agent learns to choose well when rewards are delayed and the environment is only partly known. The table of contents makes that arc explicit, moving from the introductory setup and tic-tac-toe example into bandits, finite Markov decision processes, and the core questions of prediction and control. The book then builds the field methodically: dynamic programming, Monte Carlo methods, temporal-difference learning, and n-step bootstrapping each get their own treatment. Along the way, it covers policy evaluation and improvement, exploration strategies, importance sampling, Sarsa, Q-learning, expected Sarsa, and double learning, with the later chapters extending the same ideas to more advanced algorithmic variants. That structure gives the reader more than a survey, it gives a working map of the subject from first principles to practical algorithms. The result is a rigorous reference for understanding how reinforcement learning methods are derived, how they differ, and why they work, especially for readers who want the classic foundation of the field rather than a lightweight introduction.