A Semi-Supervised Clustering Approach For Graph Learning with Neural Networks

Chester Tan, Anatol Eugen Wegner, Ingo Scholtes

Keywords: Semi-Supervised Clustering, Node Classification, Graph Clustering, Stochastic Block Model, Graph Neural Network, Transformer, MLP

Abstract:

We propose a semi-supervised approach that combines any unsupervised clustering objective and supervised objective for end-to-end training any neural networks to improve node classification in attributed graphs, particularly when training labels are sparse.
Our framework formulates node classification as semi-supervised inference of neural network models of attributed graphs with cluster structure.
We use this framework to understand how neural networks for graph clustering can jointly cluster node attributes and graph structure, despite graph clustering objectives explicitly considering only graph structure and cluster assignments.
Our framework also enables neural network architectures such as transformers and multilayer perceptrons to learn on graphs without positional encodings and without spectral or message passing layers found in graph neural networks.
We evaluate our framework on six real-world attributed graph datasets.

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.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

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: 6861