Agents continuously communicate with the environment, receive the communication messages from others, give meaningful feedback without centralized learning.

Use Causal Structured Model (CSM) for Markov games → frames MA communication problem as a MA causal model → represents the MDP as directed acyclic graph (DAG).

Communication framework: encode current observations → communication vectors.

@ARTICLE{9761961, author={Wang, Han and Yu, Yang and Jiang, Yuan}, journal={IEEE Transactions on Neural Networks and Learning Systems}, title={Fully Decentralized Multiagent Communication via Causal Inference}, year={2023}, month={Dec.}, volume={34}, number={12}, pages={10193-10202}, }

Shield Decentralization for Safe Multi-Agent Reinforcement Learning

NIPS

2022

Centralized shield requires agents to communicate with each other.

Design shielding strategy for MARL.

The decentralized shields enable each agent to act independently, without any coordination after shield synthesis.

@inproceedings{NEURIPS2022_57444e14, author = {Melcer, Daniel and Amato, Christopher and Tripakis, Stavros}, booktitle = NIPS, editor = {S. Koyejo and S. Mohamed and A. Agarwal and D. Belgrave and K. Cho and A. Oh}, title = {Shield Decentralization for Safe Multi-Agent Reinforcement Learning}, month={Dec.}, year = {2022} }

Decentralized Policy Gradient Descent Ascent for Safe Multi-Agent Reinforcement Learning

AAAI

2021

Introduces safe decentralized policy gradient → deal with MARL over a graph.

When the graph is fully connected → safe Dec-PG will reduces to a centralized algorithm.

Taming Communication and Sample Complexities in Decentralized Policy Evaluation for Cooperative Multi-Agent Reinforcement Learning

NIPS

2021

Decentralized policy evaluations (PE) by Nonlinear function approximations.

Low communication complexity.

Finite-Sample Analysis for Decentralized Batch Multi-agent Reinforcement Learning With Networked Agents

There are no rows in this table

⁠

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.