FedMABench
收藏资源简介:
FedMABench是由浙江大学上海人工智能实验室创建的,包含6个数据集,旨在为联邦训练和评估移动代理提供标准化基准。该数据集涵盖了30多个子集,8种联邦算法,10多种基础模型,以及跨5个类别的800多个应用程序,提供了一个全面的框架,用于评估不同环境下的移动代理。数据集基于Android Control和Android in the Wild数据集,针对移动使用的多样性,设计了不同类型的数据集,包括同质和异质场景,以促进进一步研究。
FedMABench was developed by the Shanghai AI Laboratory of Zhejiang University. It comprises 6 datasets, aiming to provide a standardized benchmark for federated training and evaluation of mobile agents. The benchmark encompasses over 30 subsets, 8 federated learning algorithms, more than 10 foundational models, and over 800 applications spanning 5 categories, offering a comprehensive framework for evaluating mobile agents across various environments. Grounded in the Android Control and Android in the Wild datasets, FedMABench designs diverse dataset configurations including homogeneous and heterogeneous scenarios to accommodate the diversity of mobile usage patterns, so as to facilitate further research.




