Lifelong Sequential Knowledge Editing without Model Degradation
- URL: http://arxiv.org/abs/2502.01636v1
- Date: Mon, 03 Feb 2025 18:59:14 GMT
- Title: Lifelong Sequential Knowledge Editing without Model Degradation
- Authors: Akshat Gupta, Phudish Prateepamornkul, Maochuan Lu, Ahmed Alaa, Thomas Hartvigsen, Gopala Anumanchipalli,
- Abstract summary: We show that locate-then-edit knowledge editing methods lead to overfitting on the edited facts.
We show that continuous knowledge editing using these methods leads to disproportionate growth in the norm of the edited matrix.
We present ENCORE - Early stopping and Norm-Constrained Robust knowledge Editing.
- Score: 11.14177136208272
- License:
- Abstract: 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.
Related papers
- AnyEdit: Edit Any Knowledge Encoded in Language Models [69.30638272162267]
We propose AnyEdit, a new autoregressive editing paradigm for large language models (LLMs)
It decomposes long-form knowledge into sequential chunks and iteratively edits the key token in each chunk, ensuring consistent and accurate outputs.
It outperforms strong baselines by 21.5% on benchmarks including UnKEBench, AKEW, and our new EditEverything dataset for long-form diverse-formatted knowledge.
arXiv Detail & Related papers (2025-02-08T16:18:37Z) - Reasons and Solutions for the Decline in Model Performance after Editing [17.756172082400163]
This paper explores the reasons for the performance decline of the edited model and optimize the editing method.
The performance of the editing model is mainly affected by the diversity of editing targets and sequence length.
In order to improve the performance of the editing model, this paper proposes a Dump for Sequence (D4S) method.
arXiv Detail & Related papers (2024-10-31T11:49:44Z) - O-Edit: Orthogonal Subspace Editing for Language Model Sequential Editing [0.0]
Large language models (LLMs) acquire knowledge during pre-training, but over time, this knowledge may become incorrect or outdated, necessitating updates after training.
We propose Orthogonal Subspace Editing, O-Edit. This algorithmizes the direction of each knowledge update, minimizing interference between successive updates and reducing the impact of new updates on unrelated knowledge.
It can perform thousands of edits on mainstream LLMs, achieving an average performance improvement that is 4.2 times better than existing methods while effectively preserving the model's performance on downstream tasks, all with minimal additional parameter overhead.
arXiv Detail & Related papers (2024-10-15T10:16:45Z) - AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models [65.93240009586351]
Large language models (LLMs) often exhibit hallucinations due to incorrect or outdated knowledge.
We introduce AlphaEdit, a novel solution that projects perturbation onto the null space of the preserved knowledge before applying it to the parameters.
We theoretically prove that this projection ensures the output of post-edited LLMs remains unchanged when queried about the preserved knowledge.
arXiv Detail & Related papers (2024-10-03T10:06:27Z) - Outdated Issue Aware Decoding for Reasoning Questions on Edited Knowledge [93.54427119091174]
We propose outDated ISsue aware deCOding to enhance the performance of edited models on reasoning questions.
We capture the difference in the probability distribution between the original and edited models.
We amplify the difference of the token prediction in the edited model to alleviate the outdated issue.
arXiv Detail & Related papers (2024-06-05T03:00:15Z) - WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models [78.22291694903659]
Large language models (LLMs) need knowledge updates to meet the ever-growing world facts and correct the hallucinated responses.
Where the updated knowledge resides in memories is a fundamental question for model editing.
We propose WISE to bridge the gap between memories.
arXiv Detail & Related papers (2024-05-23T16:35:52Z) - Rebuilding ROME : Resolving Model Collapse during Sequential Model Editing [2.569159339315845]
We show that disabling edits are an artifact of irregularities in the implementation of Rank-One Model Editing (ROME)
We provide a more stable implementation ROME, which we call r-ROME, and show that model collapse is no longer observed when making large scale sequential edits with r-ROME.
arXiv Detail & Related papers (2024-03-11T21:33:05Z) - WilKE: Wise-Layer Knowledge Editor for Lifelong Knowledge Editing [19.357663224043534]
This study reveals a performance degradation encountered by knowledge editing in lifelong editing.
We introduce a knowledge editing approach named Wise-Layer Knowledge Editor (WilKE)
WilKE selects editing layer based on the pattern matching degree of editing knowledge across different layers in language models.
arXiv Detail & Related papers (2024-02-16T05:29:59Z) - The Butterfly Effect of Model Editing: Few Edits Can Trigger Large Language Models Collapse [58.0132400208411]
Even a single edit can trigger model collapse, manifesting as significant performance degradation in various benchmark tasks.
benchmarking Large Language Models after each edit is impractically time-consuming and resource-intensive.
We have utilized GPT-3.5 to develop a new dataset, HardEdit, based on hard cases.
arXiv Detail & Related papers (2024-02-15T01:50:38Z) - Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue [122.20016030723043]
We evaluate the side effects of model editing on large language models (LLMs)
Our analysis reveals that the side effects are caused by model editing altering the original model weights excessively.
To mitigate this, a method named RECT is proposed to regularize the edit update weights.
arXiv Detail & Related papers (2024-01-09T18:03:15Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.