Reinforcement Learning

Course dates – HT – week beginning Monday 18th February 2019 – for Year 1 students
Shimon Whiteson


Reinforcement learning is the process of learning by trial and error to maximise a reward signal, and is applicable to a wide range of tasks from robotics and game playing to information retrieval, recommender systems, and question answering tasks.  This short course gives an introduction the foundational concepts and surveys some of the exciting recent developments in deep reinforcement learning.

  • understand the foundational concepts of decision-theoretic planning and reinforcement learning
  • get acquainted with some of the recent developments in deep reinforcement learning
  • Markov decision processes and value functions
  • Model-free reinforcement learning
  • Linear and nonlinear value function approximation
  • Deep reinforcement learning
  • Policy gradient methods
  • Model-based reinforcement learning
  • Partially observable Markov decision processes
  • Cooperative multi-agent reinforcement learning


  • Calculus, linear algebra, probability and statistics, programming, machine learning, neural networks
Other Sources
  • Sutton & Barto, 'Reinforcement Learning: An Introduction’, Second Edition, 2018 (pdf free online), Chapters 3-6, 8-11, and 13