Does Structural Information have been Fully Exploited in Graph Data?

Xin Lai, Yang Aron Liu, Rui Qian, Yong Lin, Qiwei Ye

Keywords: Transformer, Discrete Ricci Curvature, Structural Information

TL;DR: We propose a topology-aware graph-based architecture, termed as Curvphormer, by leveraging a geometric notion, i.e., discrete Ricci curvature.

Abstract:

In real world, graph-structured data is pervasive, operating as an abstraction of data containing nodes and interactions between nodes. There are numerous ways dedicated to excavating structure information overtly or implicitly, but whether structural information has been adequately exploited remains an unanswered question. We offer Curvphormer, a curvature-based topology-aware Graphormer that integrates Discrete Ricci Curvature (DRC) into a powerful graph-based Transformer architecture to construct a more expressive graph-based model. This work employs DRC, a geometric descriptor, to reveal additional structural information. We intuitively characterize how our model can make better use of the topological information in graph data, and extract desired structural information, such as inherent community structure in graphs with homogeneous information. We conduct extensive experiments on different scaled datasets, such asg PCQM4M-LSC, ZINC and MolHIV, and achieve remarkable performance gain on various graph-level tasks and finetune tasks. Codes
will be released upon acceptance.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning