Reinforcement learning and Optimisation (DSA Reading Group) - PART 2
Reinforcement learning (RL) is a branch of machine learning which has the potential to be used to solve problems in optimisation. Current state-of-the-art RL algorithms are capable of successfully dealing with problems which involve extremely large numbers of possible states, however it is often difficult or impossible to extend these RL algorithms to problems which also involve extremely large numbers of actions. I will talk about two algorithms recently proposed which attempt to deal with extremely large action spaces: these are described in "Deep reinforcement learning in large discrete action spaces" by Dulac-Arnold et al, and "Learn What Not to Learn: Action Elimination with Deep Reinforcement Learning" by Zahavy et al). Both papers were coauthored by researchers from Google.
Mr Edward Barker,