BREC

BREC数据集由北京大学人工智能研究所的Yanbo Wang和Muhan Zhang创建，包含400对非同构图，分为基本、规则、扩展和CFI四个主要类别。数据集难度高达4-WL不可区分，粒度精细，能够比较1-WL到3-WL的模型，规模达到400对。创建过程中，通过精心挑选和设计图结构，确保了数据集的高质量和多样性。BREC数据集主要用于测试未来GNN的表达能力，解决图神经网络在区分非同构图方面的局限性。

The BREC dataset was created by Yanbo Wang and Muhan Zhang from the Institute of Artificial Intelligence at Peking University. It contains 400 pairs of non-isomorphic graphs, divided into four main categories: Basic, Regular, Extended, and CFI. Featuring 4-WL indistinguishability and fine granularity, the dataset enables comparative evaluation of models ranging from 1-WL to 3-WL, with a total scale of 400 pairs. During its creation, the graph structures were carefully selected and designed to ensure high quality and diversity of the dataset. The BREC dataset is primarily used to test the expressive capabilities of future graph neural networks (GNNs), addressing the limitations of GNNs in distinguishing non-isomorphic graphs.