Efficient Communication in Multi-Agent Reinforcement Learning via Variance Based Control

Authors: Sai Qian Zhang, Qi Zhang, Jieyu Lin

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our evaluation using multiple MARL benchmarks indicates that our method achieves 2 10 lower in communication overhead than state-of-the-art MARL algorithms, while allowing agents to achieve better overall performance.
Researcher Affiliation Collaboration Sai Qian Zhang Harvard University Qi Zhang Amazon Inc. Jieyu Lin University of Toronto
Pseudocode Yes Algorithm 1: Communication protocol at agent i
Open Source Code Yes The code is available at https://github.com/saizhang0218/VBC.
Open Datasets Yes For evaluation, we test VBC on several MARL benchmarks, including Star Craft Multi-Agent Challenge [15], Cooperative Navigation (CN) [10] and Predator-prey (PP) [8].
Dataset Splits No The paper describes training duration (e.g., '2 million and 4 million episodes') and test episodes ('20 test episodes'), but does not explicitly specify train/validation/test dataset splits, percentages, or sample counts.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory).
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes For hyperparameters used by VBC (i.e., λ used in equation (1), δ1andδ2 in Algorithm 1), we first search for a coarse parameter range based on random trial, experience and message statistics. We then perform a random search within a smaller hyperparameter space. Best selections are shown in the legend of each figure.