Conflict-Resolving and Sharpness-Aware Minimization for Generalized Knowledge Editing with Multiple Updates
Duy Nguyen, Hanqi Xiao, Archiki Prasad, Elias Stengel-Eskin, Hyunji Lee, Mohit Bansal · 2026
Read Conflict-Resolving and Sharpness-Aware Minimization for Generalized Knowledge Editing with Multiple Updates by Duy Nguyen, Hanqi Xiao, Archiki Prasad, Elias Stengel-Eskin, Hyunji Lee, Mohit Bansal (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-2602-03696.txt.
Description
Editing a language model's facts sounds surgical, but repeated edits often conflict, generalize poorly, or destabilize the model. Duy Nguyen and 5 co-authors present CoRSA in a 2026 paper, a training framework for knowledge editing with multiple updates. It is parameter-efficient and tries to solve the generalization, stability, and conflict problems together.
CoRSA minimizes loss curvature, a sharpness-aware move that steadies behavior across successive updates. It also maximizes the margin between new and prior knowledge to resolve conflicts directly. On three fact-editing benchmarks it improves generalization well beyond LoRA and editing baselines. With many updates it keeps efficacy high while cutting catastrophic forgetting, and it transfers to code editing as well.
Catalog record
| CREATOR | Duy Nguyen, Hanqi Xiao, Archiki Prasad, Elias Stengel-Eskin, Hyunji Lee, Mohit Bansal |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03696