RL in Discrete Optimization

The perception of deep reinforcement learning (RL) is that it is a tool to be used mostly for playing games. This is not surprising given the fact that, historically, the first success in the field was achieved on the Atari game suite by DeepMind in 2015 (https://deepmind.com/research/dqn/). The Atari benchmark suite turned out to be very successful for RL problems and, even now, lots of research papers use it to demonstrate the efficiency of their methods. As the RL field progresses, the classic 53 Atari games continue to become less and less challenging (at the time of writing, almost all the games have been solved with superhuman accuracy) and researchers are turning to more complex games, like StarCraft and Dota 2.

This perception, which is especially prevalent in the media, is something that I’ve tried to counterbalance in this book by accompanying Atari games with examples from other domains, including stock trading and natural...