FedMABench

该数据集名为FedMABench，是首个专为异构场景设计的、用于联邦训练和评估移动代理的基准测试。它包含了6个数据集，拥有超过30个子集，为在多样化环境中评估移动代理提供了一个全面的框架。这些数据集旨在促进对联邦学习算法的实验研究，并在不同的移动应用和用户场景中评估性能。规模上，它跨越了5个类别，包含超过800个应用，形成了6个数据集和30多个子集的丰富组合。其任务是对移动代理的联邦学习算法进行评估。

This dataset, named FedMABench, is the first benchmark exclusively designed for heterogeneous scenarios to support federated training and evaluation of mobile agents. It comprises 6 datasets with over 30 subsets, providing a comprehensive framework for evaluating mobile agents in diverse environments. This benchmark aims to facilitate experimental research on federated learning algorithms and assess their performance across various mobile applications and user scenarios. In terms of scale, it spans 5 categories and includes more than 800 applications, forming a rich portfolio of 6 datasets and over 30 subsets. Its core objective is to evaluate federated learning algorithms for mobile agents.