多智能通信消息聚合优化数据集

面向为解决多智能体协同任务中，由于通信消息同质化导致的协同效率变差问题。本数据集记录了多智能体强化学习通信策略模型训练过程中通信消息的差异量化指标，并保存了训练后具有差异性的多智能体通信策略模型。数据集以星际争霸多智能体任务环境（SMAC，评价多智能体强化学习工作的通用基准）为基础，包含多个智能体在SMAC任务中的训练日志数据和模型。训练日志数据使用tensorboard记录，模型使用pytorch保存。数据量300.00MB。

To address the problem of deteriorated collaborative efficiency induced by homogeneous communication messages in multi-agent collaborative tasks, this dataset captures the difference quantification metrics of communication messages during the training of multi-agent reinforcement learning-based communication strategy models, and preserves the trained multi-agent communication strategy models with distinguishable differences. Built upon the StarCraft Multi-Agent Challenge (SMAC), a widely-used universal benchmark for evaluating multi-agent reinforcement learning research, this dataset includes training log data and models of multiple agents across SMAC tasks. The training log data is logged via TensorBoard, while the models are saved using PyTorch. The total size of this dataset is 300.00 MB.

搜集汇总

背景与挑战

背景概述

该数据集旨在解决多智能体协同任务中通信消息同质化导致的效率下降问题，记录了多智能体强化学习通信策略训练过程中的消息差异量化指标和最终模型。数据基于星际争霸多智能体任务环境（SMAC），包含训练日志和模型文件，总数据量约为300MB。