Generative Risk Minimization for Out-of-Distribution Generalization on Graphs
Song Wang [email protected], Zhen Tan [email protected], Yaochen Zhu [email protected], Chuxu Zhang [email protected], Jundong Li [email protected] · 2025
Read Generative Risk Minimization for Out-of-Distribution Generalization on Graphs by Song Wang [email protected], Zhen Tan [email protected], Yaochen Zhu [email protected], Chuxu Zhang [email protected], Jundong Li [email protected] (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-2502-07968.txt.
Description
A graph classifier trained in one world often fails in another, because the background structure shifts while the true signal stays put. Song Wang and co-authors propose generating the signal part rather than carving it out. Their framework, Generative Risk Minimization, builds a fresh invariant subgraph for each input graph instead of extracting one.
Carving risks cutting away real signal or keeping junk; generation can, in principle, rebuild exactly the stable core. The snag is training without any examples of the perfect core. The authors introduce a hidden causal variable that makes the goal computable, and they prove the approach sound in theory. On real graph datasets, at both node and whole-graph level, their method beat the extraction-based rivals.
Catalog record
| CREATOR | Song Wang [email protected], Zhen Tan [email protected], Yaochen Zhu [email protected], Chuxu Zhang [email protected], Jundong Li [email protected] |
|---|---|
| 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/2502-07968