Reproduce “An Axiomatic Framework for Reinforcement Learning in Open Multi-Agent Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jianhong Wang, Yang Li, Samuel Kaski, Jonathan Lawry
The authors develop a value-learning framework for open teams, where only some agents are controlled and teammates may change. It maps Efficiency, Symmetry, and Linearity axioms to reward shaping, permutation-consistent structures, and multi-step temporal-difference learning. It recovers Shapley values in theory; in tests, Shapley methods lead on unseen conventions except 10v11, while Banzhaf methods match or outperform them in most unseen-type tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.