Rethinking Individual Global Max in Cooperative Multi-Agent Reinforcement Learning

Authors: Yitian Hong, Yaochu Jin, Yang Tang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed strategy is theoretically proved and empirically verified on the Star Craft Multi-Agent Challenge benchmark problem with zero sight view. The results also confirm that the proposed method outperforms state-of-the-art IGM-based approaches.
Researcher Affiliation Academia Yitian Hong East China University of Science and Technology ythong1314@mail.ecust.edu.cn Yaochu Jin Bielefeld University yaochu.jin@uni-bielefeld.de Yang Tang East China University of Science and Technology yangtang@ecust.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes The code of the proposed algorithm can be downloaded at *. *https://github.com/momo-xiaoyi/pymarl_HDA
Open Datasets Yes To rigorously investigate the performance of the proposed learning strategy, we adopt the Star Craft II benchmark task as the test problem. [...] The Star Craft Multi-Agent Challenge benchmark problem
Dataset Splits No The paper refers to the Star Craft II benchmark, but does not explicitly provide specific train/validation/test dataset splits, percentages, or sample counts. It mentions "implementation details and experimental settings can be found in Appendices C and D" but these appendices do not detail dataset splitting for reproduction.
Hardware Specification No No specific hardware details (e.g., GPU models, CPU models, memory) used for running the experiments are provided in the paper.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes For fair evaluations, the hyper-parameters of all algorithms under comparison as well as the optimizers, are the same, and the experimental results are presented with the average performance with 25-75% percentile. Moreover, the presented curves are smoothed by a moving average filter with its window size being set to 5 for better visualization. The implementation details and experimental settings can be found in Appendices C and D.