Wander: A Unified Graph Foundation Model for Cross-Task and Cross-Modal Learning

Researchers have introduced "Wander," a novel graph foundation model designed to generalize across diverse graph modalities, feature spaces, relational schemas, and prediction tasks. Unlike previous approaches that are limited to specific graph types or tasks, Wander utilizes a unified interface based on random walks. This allows a single pretrained model to operate effectively on homogeneous and multi-relational graphs, even with varying features and labels. Empirically, Wander has demonstrated state-of-the-art or highly competitive results in node classification, homogeneous link prediction, and knowledge-graph link prediction, showcasing its ability to transfer learning and compose capabilities across different settings. AI

IMPACT This research could enable more versatile and efficient graph-based AI applications by allowing a single model to handle diverse data structures and tasks.

RANK_REASON The cluster contains a research paper detailing a new model architecture and its empirical evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv Recommender
• Influence Flower
• Litmaps
• Louis Tichelman
• ScienceCast
• scite Smart Citations
• Wander

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →