Energy-Regularized Sequential Model Editing on Hyperspheres
Qingyuan Liu, Jia-Chen Gu, Yunzhi Yao, Hong Wang, and Nanyun Peng, in Proceedings of the International Conference on Learning Representations (ICLR), 2026.
Download the full text
Abstract
The paper aims to understand and reduce the performance degradation caused by sequential editing. The authors use Hyperspherical Energy (HE) to quantify neuron uniformity during editing and investigate its correlation with editing performance. They propose SPHERE, an HE-driven regularization strategy that stabilizes neuron weight distributions, thereby preserving prior knowledge and enabling reliable sequential updates.
Bib Entry
@inproceedings{liu2026energy,
title = {Energy-Regularized Sequential Model Editing on Hyperspheres},
author = {Liu, Qingyuan and Gu, Jia-Chen and Yao, Yunzhi and Wang, Hong and Peng, Nanyun},
booktitle = {Proceedings of the International Conference on Learning Representations (ICLR)},
year = {2026}
}