GFMate: Empowering Graph Foundation Models with Test-time Prompt Tuning

Researchers introduced GFMate, a pre-training-agnostic test-time prompt tuning method for graph foundation models.

GFMate combines centroid and layer prompts with a complementary learning objective to adapt to target domains using unlabeled data.

Across 12 benchmark datasets, the method improved performance by up to 30.63%.

The approach addresses a major weakness of graph prompt methods by reducing reliance on source-domain-specific prompts and boosting cross-domain generalization.