benchmark_graphinference

本研究引入了名为‘benchmark_graphinference’的数据集，由IMT Atlantique和University of Rochester共同创建。该数据集包含图像、音频、文本和交通信息等多种类型的信号，旨在评估图推理方法的有效性。数据集的创建涉及从预训练的深度神经网络中提取特征，以及使用实际交通数据进行信号去噪。此数据集主要应用于图推理算法的基准测试，特别是在无监督聚类、半监督分类和图信号去噪等领域，以解决图结构从数据中推断的挑战。

This study introduces a dataset named 'benchmark_graphinference', which was jointly developed by IMT Atlantique and the University of Rochester. This dataset encompasses various types of signals including images, audio, text, and traffic information, and aims to evaluate the effectiveness of graph inference methods. The creation of this dataset involves extracting features from pre-trained deep neural networks and performing signal denoising using real-world traffic data. This dataset is primarily utilized for benchmarking graph inference algorithms, particularly in domains such as unsupervised clustering, semi-supervised classification, and graph signal denoising, to address the challenge of inferring graph structures from data.