Equivalence is All: A Unified View for Self-supervised Graph Learning

Yejiang Wang, Yuhai Zhao, Zhengkui Wang, Ling Li, Jiapu Wang, Fangting Li, Miaomiao Huang, Shirui Pan, Xingwei Wang

International Conference on Machine Learning 2025 · Oral

This talk introduces **GALE (Graph Automorphic Equivalence Learning)**, a novel self-supervised learning framework for graph-structured data that unifies and leverages the fundamental concept of node equivalence. Presented by one of the co-authors on behalf of the first author, Yejiang Wang, due to unforeseen circumstances, the work highlights that real-world networks inherently possess various forms of node equivalence—from perfect structural symmetries in molecules to shared functional roles in social or biological networks. GALE posits that an ideal graph representation model should learn embeddings that reflect these equivalences, pushing equivalent nodes closer in the embedding space while separating non-equivalent ones.

AI review

GALE proposes a self-supervised graph learning framework grounded in node equivalence — automorphic and attribute-based — using a dual-branch architecture and a contrastive-style equivalence loss. The central organizing idea is clean and has genuine explanatory appeal: reframing existing graph SSL methods (GCL, MPNNs, graph transformers) through the lens of equivalence classes is a useful unifying move. The empirical results appear competitive, and the ablation study on GCL — showing that positive pairing dominates augmentation — is the most interesting finding in the talk. However, as…