XQMIX: Extended QMix for StarCraft Multi-Agent Challenge

Mohammad Mahdi Rahimi

Technical report, Korea Advanced Institute of Science and Technology (KAIST) · 2021 · DOI: 10.13140/RG.2.2.23575.91040

Abstract

This report describes improvements to decentralized multi-agent learning for the StarCraft II Multi-Agent Challenge. The principal additions are multi-step learning and noisy networks. Other evaluated changes include the optimizer, learning-rate decay schedule, loss functions, and regularization. The report analyzes which changes improve upon the QMIX baseline and where additional hyperparameter tuning is required.

Canonical citation

Rahimi, M. M. (2021). XQMIX: Extended QMix for StarCraft Multi-Agent Challenge. Technical report, KAIST. https://doi.org/10.13140/RG.2.2.23575.91040

@techreport{rahimi2021xqmix,
  title = {XQMIX: Extended QMix for StarCraft Multi-Agent Challenge},
  author = {Rahimi, Mohammad Mahdi},
  institution = {Korea Advanced Institute of Science and Technology},
  year = {2021},
  doi = {10.13140/RG.2.2.23575.91040}
}

Archive note: this author-hosted PDF preserves the visible pages of the official version. Only embedded document metadata was normalized so title and author fields match the title page.