An overview of the Arcade Learning Environment (ALE)
- 11:00 21st November 2019 ( week 6, Michaelmas Term 2019 )Lecture Theatre B, Wolfson Building
The Arcade Learning Environment (ALE), originally introduced by Bellemare et al. (2013), is a central benchmark for evaluating Reinforcement Learning (RL) algorithms. It was notably employed by Mnih et al. (2013) to demonstrate the abilities of their Deep Q-Network, and years later continues to be widely used. Composed of Atari 2600 games, it presents a diverse set of challenges of varying difficulty, helping distinguish the strengths and weaknesses of RL techniques. Its popularity makes it an important benchmark for those hoping to work on RL techniques. In this talk, I will give an introduction to the ALE games and will explain subtle but important implementation details for those who might want to use the benchmark.