Beyond the Covariance Trap: Unlocking Generalization in Same-Subject Knowledge Editing for Large Language Models

TL;DR

This paper identifies the geometric causes of generalization failure in same-subject knowledge editing of LLMs and proposes RoSE, a method that enhances robustness and instruction-following by geometric alignment and hierarchical knowledge integration.

Contribution

It introduces RoSE, a novel approach that addresses geometric and optimization issues in knowledge editing, improving model generalization and robustness.

Findings

RoSE significantly improves instruction-following accuracy.

It reduces the covariance trap effect in knowledge editing.

The method enhances the robustness of LLMs in interactive scenarios.

Abstract

While locate-then-edit knowledge editing efficiently updates knowledge encoded within Large Language Models (LLMs), a critical generalization failure mode emerges in the practical same-subject knowledge editing scenario: models fail to recall the updated knowledge when following user instructions, despite successfully recalling it in the original edited form. This paper identifies the geometric root of this generalization collapse as a fundamental conflict where the inner activation drifts induced by prompt variations exceed the model's geometric tolerance for generalization after editing. We attribute this instability to a dual pathology: (1) The joint optimization with orthogonal gradients collapses solutions into sharp minima with narrow stability, and (2) the standard covariance constraint paradoxically acts as a Covariance Trap that amplifies input perturbations. To resolve this, we introduce RoSE (Robust Same-subject Editing), which employs Isotropic Geometric Alignment to minimize representational deviation and Hierarchical Knowledge Integration to smooth the optimization landscape. Extensive experiments demonstrate that RoSE significantly improves instruction-following capabilities, laying the foundation for robust interactive parametric memory of LLM agents.