Revealing and Mitigating Over-Attention in Knowledge Editing

TL;DR

This paper identifies that over-attention in language models causes knowledge editing errors and proposes SADR, a regularization method to restrict attention shifts, effectively reducing such errors across multiple models.

Contribution

The paper introduces SADR, a novel regularization technique that mitigates over-attention during knowledge editing, improving model reliability and preserving pre-existing knowledge.

Findings

SADR significantly reduces Specificity Failure in knowledge editing.

The method is effective across five large language models.

SADR maintains model performance while preventing attention over-focus.

Abstract

Large Language Models have demonstrated superior performance across a wide range of tasks, but they still exhibit undesirable errors due to incorrect knowledge learned from the training data. To avoid this, knowledge editing methods emerged to precisely edit the specific model knowledge via efficiently modifying a very small percentage of parameters. % However, those methods can lead to the problem of Specificity Failure: when the content related to the edited knowledge occurs in the context, it can inadvertently corrupt other pre-existing knowledge. However, those methods can lead to the problem of Specificity Failure, where the existing knowledge and capabilities are severely degraded due to editing. Our preliminary indicates that Specificity Failure primarily stems from the model's attention heads assigning excessive attention scores to entities related to the edited knowledge, thereby unduly focusing on specific snippets within the context, which we denote as the Attention Drift phenomenon. To mitigate such Attention Drift issue, we introduce a simple yet effective method Selective Attention Drift Restriction}(SADR), which introduces an additional regularization term during the knowledge editing process to restrict changes in the attention weight distribution, thereby preventing undue focus on the edited entity. Experiments on five frequently used strong LLMs demonstrate the effectiveness of our method, where SADR can significantly mitigate Specificity Failure in the predominant knowledge editing tasks.