We introduce a new mathematical model of multi-agent reinforcement learning, the Multi-Agent Informational Learning Processor "MAILP" model. The model is based on the notion that agents have policies for a certain amount of information, models how this information iteratively evolves and propagates through many agents. This model is very general, and the only meaningful assumption made is that learning for individual agents progressively slows over time.
@article{arxiv.2006.06870,
title = {Multi-Agent Informational Learning Processes},
author = {J. K. Terry and Nathaniel Grammel},
journal= {arXiv preprint arXiv:2006.06870},
year = {2021}
}
Comments
We are withdrawing this paper as section 2.1.1 implicitly assumes information gain at all points is homogenous. A researcher has provided us an example showing that this assumption causes our model to make unexpected and pathological predictions, and we are aware of now way to remove this assumption from our work