Graph Out-of-Distribution Detection via Test-Time Calibration with Dual Dynamic Dictionaries
Yue Hou, Ruomei Liu , Yingke Su, Junran Wu, Ke Xu · 2025
Read Graph Out-of-Distribution Detection via Test-Time Calibration with Dual Dynamic Dictionaries by Yue Hou, Ruomei Liu , Yingke Su, Junran Wu, Ke Xu (2025) 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-2511-13541.txt.
Description
Yue Hou and 4 co-authors propose BaCa, a test-time method for spotting out-of-distribution graphs without retraining. Detectors usually learn only from in-distribution data, never seeing true outliers, so their sense of the boundary stays weak. Graph data also hide several latent factors that shape their structure.
BaCa estimates graphons from test samples alone and mixes them to create diverse boundary-like graphs, avoiding any outside outlier set. Two dynamic dictionaries, kept by priority queues and attention, capture in- and out-distribution patterns and recalibrate the outlier score. Broad tests on real datasets showed clear gains over leading methods, with no fine-tuning of the base model.
Catalog record
| CREATOR | Yue Hou, Ruomei Liu , Yingke Su, Junran Wu, Ke Xu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-13541