MARL Chapter 9.9: Population-Based Training

Previous Meeting Recording

This meeting will continue the material from Chapter 9 in Multi-Agent Reinforcement Learning: Foundations and Modern Approaches. In the last meeting we started the discussion of population based training and the double oracle algorithm in the context of zero-sum games. This time, we will dive deeper into the theoretical limits of the algorithm by comparing it to exact solution techniques with tabular stochastic games. We will start with the zero-sum case and minimax solutions where we can compare the oracle to value iteration. If time permits, we will also consider general sum games where we will need to use a meta solver such as WoLF-PHC to solve the metagame.

As usual you can find below links to the textbook, previous chapter notes, slides, and recordings of some of the previous meetings.

Meetup Links:
Recordings of Previous RL Meetings
Recordings of Previous MARL Meetings
Short RL Tutorials
My exercise solutions and chapter notes for Sutton-Barto
My MARL repository
Kickoff Slides which contain other links
MARL Kickoff Slides

MARL Links:
Multi-Agent Reinforcement Learning: Foundations and Modern Approaches
MARL Summer Course Videos
MARL Slides

Sutton and Barto Links:
Reinforcement Learning: An Introduction by Richard S. Sutton and Andrew G. Barto
Video lectures from a similar course