Graph as Point Set

Xiyuan Wang, Pan Li, Muhan Zhang

Primary Area: learning on graphs and other geometries & topologies

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.

Keywords: Graph Neural Network

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A new paradigm for graph representation learning: converting interconnected nodes to independent nodes and encode the point set.

Abstract: Graphs, fundamental data structures with diverse real-world applications, consist of interconnected nodes. Existing Graph Neural Networks (GNNs) have predominantly concentrated on encoding these intricate interconnections. They employ either edge-based mechanisms for guiding message passing between nodes or complex neural network architectures designed to handle adjacency matrices as inputs, resulting in a plethora of intricate designs. Departing from this conventional trajectory, this paper introduces a paradigm-shifting approach by unveiling a novel graph-to-set conversion method. This innovative technique bijectively transforms interconnected nodes into independent points, amenable to processing by a set encoder. Utilizing the Transformer, a standard set model, we facilitate point learning. Theoretically, our proposed method outperforms various existing models in terms of both short-range and long-range expressivity. Extensive experimental validation further substantiates our model's real-world performance.

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.

Submission Number: 729