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

2026-03-16

AI Summary

This work addresses the limited generalization of large language models after subject-centric knowledge editing, where updated knowledge is only accurately recalled under the exact phrasing used during editing and fails to transfer to diverse instruction-following prompts. The authors identify a “covariance trap” caused by covariance constraints that destabilize the geometry of the activation space. To mitigate this, they propose a robust knowledge editing method that minimizes representational drift through isotropic geometric alignment and smooths the optimization landscape via a hierarchical knowledge fusion mechanism, thereby avoiding sharp minima. This approach substantially enhances the model’s ability to generalize edited knowledge across varied prompts, offering a novel pathway toward reliable, interactive parametric memory.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• knowledge editing
• generalization
• large language models
• same-subject editing
• instruction following

Innovation

• Same-subject Knowledge Editing
• Generalization Collapse
• Covariance Trap
• Isotropic Geometric Alignment
• Hierarchical Knowledge Integration