高时效无线联邦学习策略第三方测试数据

该数据集内容为：对所提出的高时效无线联邦学习场景和策略进行软件仿真，记录场景参数设置，并对训练过程中的模型准确度、网络负载和训练时延进行三方测试。仿真参数和测试时所得的测量数据形成本数据集。采集方案：模型训练使用网络下载的权威开源数据集MNIST；使用python编写仿真程序模拟无线联邦学习过程，并使用matlab求解最优的梯度压缩和训练数据量选择策略，记录模型训练的准确度、时延和网络负载，并保存在表格文件中。
采集地点：泰尔实验室
采集时间：2023年6月14日
设备情况：使用主机完成仿真，其配置为Intel(R)Core(TM)i7-107002.9GHzCPU,16GBRAM,256GBSSD。

该数据集内含一个数据说明文件与一个三方测试报告。数据集包含一个文件夹、一个数据集说明文件.docx和一个三方测试报告，文件夹名称为数据集实体文件，数据集实体文件文件夹中为数据内容，包含两个数据文件和一个文档，容量约为6.24MB。

This dataset is compiled from software simulations of the proposed high-efficiency wireless federated learning scenarios and strategies, as well as measured data collected during three-party tests on model accuracy, network load, and training latency throughout the training process. The scene parameter settings recorded during the simulation and the measurement data obtained from the tests form this dataset.

Collection protocol: The model training uses the authoritative open-source MNIST dataset downloaded from the internet. A simulation program written in Python is employed to emulate the wireless federated learning process, while MATLAB is used to solve the optimal gradient compression and training data volume selection strategies. The model training accuracy, latency, and network load are recorded and saved in spreadsheet files.

Collection location: Tai'er Laboratory
Collection time: June 14, 2023
Equipment configuration: The simulation was completed on a host computer with the following specifications: Intel(R) Core(TM) i7-10700 2.9GHz CPU, 16GB RAM, and 256GB SSD.

This dataset includes a data description file and a three-party test report. Specifically, the dataset contains a folder named "Dataset Entity Files", a dataset description document in .docx format, and a three-party test report. The "Dataset Entity Files" folder stores the core data content, which consists of two data files and one document, with a total size of approximately 6.24 MB.

搜集汇总

背景与挑战

背景概述

该数据集针对高时效无线联邦学习策略，通过软件仿真生成第三方测试数据，记录了场景参数以及模型准确度、网络负载和训练时延的测量结果。数据包含仿真参数、测试报告和说明文件，总容量约为6.25MB，格式为xlsx、pdf和docx。