AI & ML New Capability

SoLA introduces the first reversible model editing framework that allows precise revocation of specific knowledge updates.

arXiv · March 13, 2026 · 2603.11239

Haihua Luo, Xuming Ran, Tommi Kärkkäinen, Zhonghua Chen, Jiangrong Shen, Qi Xu, Fengyu Cong

Why it matters

Previous model editing methods often suffered from catastrophic forgetting or semantic drift. This framework uses semantic routing to modularize edits, enabling 'rollback' capability by simply removing keys from the router without retraining.

From the abstract

The dynamic evolution of real-world necessitates model editing within Large Language Models. While existing methods explore modular isolation or parameter-efficient strategies, they still suffer from semantic drift or knowledge forgetting due to continual updating. To address these challenges, we propose SoLA, a Semantic routing-based LoRA framework for lifelong model editing. In SoLA, each edit is encapsulated as an independent LoRA module, which is frozen after training and mapped to input by