無模型退化的終身序列知識編輯

Lifelong Sequential Knowledge Editing without Model Degradation

February 3, 2025

作者: Akshat Gupta, Phudish Prateepamornkul, Maochuan Lu, Ahmed Alaa, Thomas Hartvigsen, Gopala Anumanchipalli

cs.AI

摘要

先前在參數修改知識編輯方面的研究表明，大規模的順序編輯會導致模型明顯退化。本文研究了這背後的原因，並將順序知識編輯擴展到10,000個順序編輯，同時保持原始模型的下游性能。我們首先展示了定位-然後-編輯知識編輯方法導致對編輯事實的過度擬合。我們還展示了使用這些方法進行連續知識編輯導致編輯矩陣範數不成比例增長。然後，我們深入探討了定位-然後-編輯方法的內部運作。我們指出範數增長是這些方法使用的隱藏技巧，它使得從編輯層產生的輸出激活更加重要。通過這種“重要性黑客”，編輯層對模型輸出做出了更大的貢獻。為了緩解這些問題，我們提出了ENCORE - 早停止和範數受限的穩健知識編輯。ENCORE 控制過度擬合和不成比例的範數增長，實現長期的順序編輯，我們能夠進行多達10,000個順序編輯而不損失下游性能。ENCORE 在Llama3-8B上比MEMIT快61％，比AlphaEdit快64％。

English

Prior work in parameter-modifying knowledge editing has shown that large-scale sequential editing leads to significant model degradation. In this paper, we study the reasons behind this and scale sequential knowledge editing to 10,000 sequential edits, while maintaining the downstream performance of the original model. We first show that locate-then-edit knowledge editing methods lead to overfitting on the edited facts. We also show that continuous knowledge editing using these methods leads to disproportionate growth in the norm of the edited matrix. We then provide a crucial insight into the inner workings of locate-then-edit methods. We show that norm-growth is a hidden trick employed by these methods that gives larger importance to the output activations produced from the edited layers. With this "importance hacking", the edited layers provide a much larger contributions to the model's output. To mitigate these issues, we present ENCORE - Early stopping and Norm-Constrained Robust knowledge Editing. ENCORE controls for overfitting and the disproportionate norm-growth to enable long-term sequential editing, where we are able to perform up to 10,000 sequential edits without loss of downstream performance. ENCORE is also 61% faster than MEMIT and 64% faster than AlphaEdit on Llama3-8B.