Graph Learning with Distributional Edge Layouts
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 networks, edge features, graph layout
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We propose Distributional Edge Layouts to capture the wide latent energy distribution of layouts, achieving remarkable performance improvement in multiple datasets.
Abstract: Graph Neural Networks (GNNs) learn from graph-structured data by passing local messages between neighboring nodes along edges on certain topological layouts. Typically, these topological layouts in modern GNNs are deterministically computed (e.g., attention-based GNNs) or locally sampled (e.g., GraphSage) under heuristic assumptions. In this paper, we for the first time pose that these layouts can be globally sampled out of steady-state graphs following Boltzmann distribution equipped with explicit physical energy, leading to more viable pairwise distance configurations in the physical world. We argue that a collection of sampled/optimized layouts can capture the wide energy distribution and better characterize the intrinsic properties of input topology, therefore easing downstream tasks. As such, we propose Distributional Edge Layouts (DELs) to serve as a complement to a variety of GNNs. DEL is a pre-processing strategy independent of subsequent GNN variants, thus being highly flexible. Experimental results demonstrate that DELs consistently and substantially improve a series of GNN baselines, achieving state-of-the-art performance on multiple datasets. DEL is open-sourced at https://anonymous.4open.science/r/DEL.
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: 595
Loading