Attention Dispersion in Dynamic Graph Transformers: Diagnosis and a Transferable Fix
Jinhao Zhang, Kangfei Zhao, Qiuhao Zeng, Long-Kai Huang · 2026
Read Attention Dispersion in Dynamic Graph Transformers: Diagnosis and a Transferable Fix by Jinhao Zhang, Kangfei Zhao, Qiuhao Zeng, Long-Kai Huang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-16112.txt.
Description
Zhang and three colleagues diagnose attention dispersion as a shared failure of dynamic graph transformers under time shift. Predictions hinge on key nodes that carry more signal than random neighbours, yet models fail to focus on them when shift blurs attention contrast. Swapping standard attention for differential attention damps shared signals and boosts distinctive ones. Added to three baselines, it helps most on high-shift data, and their DiffDyG system tops nine benchmarks.
Catalog record
| CREATOR | Jinhao Zhang, Kangfei Zhao, Qiuhao Zeng, Long-Kai Huang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 87294 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.16112