中文

通过延迟机制实现DareFightingICE智能体评估的公平性

网络与互联网体系结构 2023-12-27 v1 人工智能 性能

摘要

本文提出了一种延迟机制,以减轻gRPC框架(一种高性能、开源通用远程过程调用(RPC)框架)中不同编程语言之间的延迟差异对DareFightingICE(一个格斗游戏研究平台)中智能体性能的影响。研究发现,Java和Python之间的gRPC延迟差异会显著影响实时决策。没有延迟机制时,基于Java的智能体由于Java平台上较低的gRPC延迟而优于基于Python的智能体。然而,采用所提出的延迟机制后,基于Java和基于Python的智能体表现出相似的性能,从而实现了使用不同编程语言开发的智能体之间的公平比较。因此,这项工作强调了在DareFightingICE中开发和评估智能体时考虑gRPC延迟的至关重要性,并且所获得的见解可能扩展到其他基于gRPC的应用。

关键词

引用

@article{arxiv.2312.16010,
  title  = {Achieving Fairness in DareFightingICE Agents Evaluation Through a Delay Mechanism},
  author = {Chollakorn Nimpattanavong and Thai Van Nguyen and Ibrahim Khan and Ruck Thawonmas and Worawat Choensawat and Kingkarn Sookhanaphibarn},
  journal= {arXiv preprint arXiv:2312.16010},
  year   = {2023}
}

备注

This is an updated version of our 2023 IEEE Conference on Games paper of the same title where (a) Eqn. 3 has been revised, (b) Eqn. 5 has been added for clarity, and (c) an appendix has been added that contains the link to the source code and raw data