Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Structure-Centric Graph Foundation Model via Geometric Bases

About

Graph foundation models (GFMs) seek transferable representations across graph domains but are limited by structural heterogeneity and incompatible node feature spaces. We propose Structure-Centric Graph Foundation Models (SCGFM), which treat graph topology as the primary source of transferable knowledge. Modeling graphs as metric measure spaces, SCGFM introduces learnable geometric bases that define a shared structural coordinate system. Graphs are aligned to these bases via Gromov-Wasserstein distances, yielding structure-aligned latent representations that accommodate heterogeneous graph topologies. To address feature incompatibility, SCGFM employs a structure-aware feature re-encoding mechanism that unifies node representations without assuming a fixed feature dimensionality or requiring dataset-specific preprocessing. Experiments on graph- and node-level tasks demonstrate strong in-domain and cross-domain generalization, outperforming existing GFM approaches.

Xiaodong He, Haolan He, Ruiyi Fang, Ming Sun, Zhao Kang• 2026

Related benchmarks

TaskDatasetResultRank
Graph ClassificationNCI1
Accuracy58
658
Node ClassificationCiteseer
Accuracy43.83
503
Graph ClassificationCOLLAB
Accuracy61.93
469
Node ClassificationREDDIT
Accuracy84.17
216
Graph ClassificationBZR
Accuracy60.8
165
Graph ClassificationCOX2
Accuracy56.6
161
Node ClassificationCora
Accuracy70.55
134
Graph Classificationimdb-binary
Accuracy56.62
127
Node ClassificationComputers
Accuracy (%)57.35
62
Graph Classificationcolors3
Accuracy26.97
37
Showing 10 of 12 rows

Other info

Follow for update