Summary of Less Is More: on the Over-globalizing Problem in Graph Transformers, by Yujie Xing et al.
Less is More: on the Over-Globalizing Problem in Graph Transformers
by Yujie Xing, Xiao Wang, Yibo Li, Hai Huang, Chuan Shi
First submitted to arxiv on: 2 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper challenges the notion that Graph Transformer’s global attention mechanism always benefits graph-structured data processing. It presents empirical evidence and theoretical analysis showing an “over-globalizing” problem, where distant nodes receive excessive attention while nearby nodes with valuable information are neglected. To address this issue, the authors propose Bi-Level Global Graph Transformer with Collaborative Training (CoBFormer), which combines inter-cluster and intra-cluster Transformers to prevent over-globalization while preserving distant node extraction capabilities. The model is trained collaboratively with a theoretical guarantee for improved generalization. Extensive experiments on various graphs demonstrate CoBFormer’s effectiveness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper questions whether Graph Transformer, a powerful tool for processing graph data, always works well. It shows that Graph Transformer might actually focus too much on distant nodes and ignore important information in nearby nodes. The authors propose a new model called CoBFormer to solve this problem. CoBFormer has two parts: one looks at the big picture (distant nodes) and another focuses on small details (nearby nodes). This combination helps the model learn more effectively from graph data. |
Keywords
» Artificial intelligence » Attention » Generalization » Transformer
