New Graph Foundation Model Improves Cross-Domain Transfer

Yi Wang, Jitao Zhao, Di Jin, Dongxiao He· August 3, 2026 View original

Key takeaways

  • Graph Foundation Models aim for knowledge transfer across diverse graph domains.
  • Traditional GFMs often miss dynamic propagation patterns in graphs.
  • ProGFM learns transferable propagation relationships for adaptive aggregation.
  • This approach significantly improves cross-domain generalization performance.

Who benefits

PharmaceuticalsSocial MediaCybersecurityLogisticsFinance

Summary

This paper introduces ProGFM, a Propagation-aware Graph Foundation Model, designed to enhance knowledge transfer across diverse graph domains by learning transferable propagation relationships. Unlike previous methods, ProGFM focuses on adaptive information aggregation in unseen graph domains, demonstrating superior generalization performance.

Graph Foundation Models (GFMs) are designed to transfer knowledge across different graph domains, a significant challenge due to the lack of unified representation units in graph data. Current GFMs primarily address domain differences through feature and structure alignment, often overlooking the dynamic nature of information propagation within graphs. This research proposes ProGFM, a novel Propagation-aware Graph Foundation Model. ProGFM identifies propagation relationships between edges and feature dimensions as key transferable knowledge units. By utilizing a propagation relationship prototype bank, ProGFM learns cross-domain propagation knowledge, allowing for more adaptive information aggregation and improved generalization in previously unseen graph domains.

Why it matters

This advancement could lead to more robust and adaptable AI systems that can analyze and learn from diverse graph-structured data, such as social networks, molecular structures, or supply chains, with less need for domain-specific retraining.

How to implement this in your domain

  1. 1Explore the potential of Graph Foundation Models for tasks involving diverse graph datasets in your domain.
  2. 2Investigate how "propagation knowledge" could be leveraged to improve transfer learning in your graph-based AI applications.
  3. 3Consider adopting models that can adapt their information aggregation mechanisms to different graph structures.
  4. 4Benchmark ProGFM or similar approaches against existing graph learning methods for cross-domain generalization.

Original post by Yi Wang, Jitao Zhao, Di Jin, Dongxiao He

"arXiv:2607.28980v1 Announce Type: new Abstract: Graph Foundation Models (GFMs) have recently emerged as a promising paradigm for enabling knowledge transfer across diverse domains. Unlike traditional graph learning methods that are typically designed for in-domain settings, GFMs…"

View on X

Originally posted by Yi Wang, Jitao Zhao, Di Jin, Dongxiao He on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses