Two new research papers explore the challenges and potential of graph foundation models (GFMs). The first paper, "When Do Graph Foundation Models Transfer? A Data-Centric Theory," investigates the properties of graph domains that influence the transferability of GFMs, proposing a theoretical framework to decompose output shifts and guide data curation. The second paper, "Graph is a Substrate Across Data Modalities," introduces G-Substrate, a framework designed to organize graph structure learning across heterogeneous modalities and tasks, aiming to accumulate structural regularities rather than repeatedly reconstructing them. AI
RANK_REASON Two academic papers published on arXiv discussing theoretical aspects of graph foundation models.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
