DIB-OD: Preserving the Invariant Core for Robust Heterogeneous Graph Adaptation via Decoupled Information Bottleneck and Online Distillation *Corresponding Author: Qiudong Yu
Yang Yan, Qiuyan Wang, Tianjin Huang, Qiudong Yu *, Kexin Zhang · 2026
Read DIB-OD: Preserving the Invariant Core for Robust Heterogeneous Graph Adaptation via Decoupled Information Bottleneck and Online Distillation *Corresponding Author: Qiudong Yu by Yang Yan, Qiuyan Wang, Tianjin Huang, Qiudong Yu *, Kexin Zhang (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-2604-10882.txt.
Description
Yan and colleagues offer DIB-OD, an adaptation frame pairing a decoupled information bottleneck with online distillation. Shifts in shape and features can cause harmful transfer and wipe out reusable knowledge. A multiview teacher learns a tight code, distilled into shared core and spare residual branches. Information goals and dependence rules curb overlap between branches, while a meaning rule and a frozen teacher hold pretrained knowledge steady. Tests on seven datasets from chemical, biological, and social-network fields show gains over baselines, above all in hard cross-type transfer.
Catalog record
| CREATOR | Yang Yan, Qiuyan Wang, Tianjin Huang, Qiudong Yu *, Kexin Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36903 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.10882