WilKE: Wise-Layer Knowledge Editor for Lifelong Knowledge Editing

TL;DR

WilKE is a novel knowledge editing method for large language models that improves lifelong editing performance by selecting appropriate layers based on pattern matching, significantly reducing toxicity buildup and enhancing editing accuracy.

Contribution

WilKE introduces a layer selection strategy based on pattern matching for lifelong knowledge editing, addressing toxicity issues and outperforming existing methods.

Findings

WilKE improves editing performance by 46.2% on GPT2-XL.

WilKE improves editing performance by 67.8% on GPT-J.

WilKE effectively mitigates toxicity buildup in lifelong editing.

Abstract

Knowledge editing aims to rectify inaccuracies in large language models (LLMs) without costly retraining for outdated or erroneous knowledge. However, current knowledge editing methods primarily focus on single editing, failing to meet the requirements for lifelong editing. This study reveals a performance degradation encountered by knowledge editing in lifelong editing, characterized by toxicity buildup and toxicity flash, with the primary cause identified as pattern unmatch. We introduce a knowledge editing approach named Wise-Layer Knowledge Editor (WilKE), which selects editing layer based on the pattern matching degree of editing knowledge across different layers in language models. Experimental results demonstrate that, in lifelong editing, WilKE exhibits an average improvement of 46.2% and 67.8% on editing GPT2-XL and GPT-J relative to state-of-the-art knowledge editing methods.