Long Range Graph Benchmark (LRGB)
收藏资源简介:
Long Range Graph Benchmark (LRGB) 是由南洋理工大学的Vijay Prakash Dwivedi等人创建的一个包含5个图学习数据集的基准。这些数据集包括PascalVOC-SP, COCO-SP, PCQM-Contact, Peptides-func和Peptides-struct,旨在评估模型在处理长距离交互(LRI)任务上的性能。数据集涵盖计算机视觉和化学领域,要求模型能够理解和利用图中的长距离依赖关系以实现高性能。LRGB通过对比基线GNN和图Transformer网络,验证了捕捉长距离依赖的模型在这些任务上表现更佳。这些数据集适合用于基准测试和探索旨在捕捉LRI的MP-GNN和图Transformer架构。
Long Range Graph Benchmark (LRGB) was created by Vijay Prakash Dwivedi et al. from Nanyang Technological University. It is a benchmark containing 5 graph learning datasets, including PascalVOC-SP, COCO-SP, PCQM-Contact, Peptides-func and Peptides-struct, which aims to evaluate model performance on long-range interaction (LRI) tasks. The datasets cover the fields of computer vision and chemistry, requiring models to understand and leverage long-range dependencies in graphs to achieve high performance. LRGB has verified that models capturing long-range dependencies perform better on these tasks through comparisons between baseline GNNs and graph Transformer networks. These datasets are suitable for benchmarking and exploring MP-GNN and graph Transformer architectures designed to capture LRI.

- 1Long Range Graph Benchmark南洋理工大学, 新加坡 · 2023年



