pith. sign in

arxiv: 2602.20207 · v3 · pith:HZCTRQ5Nnew · submitted 2026-02-22 · 💻 cs.LG · cs.AI

Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis

Shrestha Datta , Hongfu Liu , Anshuman Chhabra This is my paper

Pith reviewed 2026-05-21 11:36 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords knowledge editinglarge language modelsgolden layerslayer selectiongradient analysisproxy datasetparameter update
0
0 comments X

The pith

Fixed golden layers in LLMs deliver knowledge editing performance close to the per-query optimum.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to show that large language models contain fixed golden layers whose editing results nearly match the results from choosing a different best layer for every individual query. A reader would care because current approaches often test many layers per edit, which becomes costly as models grow larger. The authors test the idea by measuring how closely golden-layer edits track the actual best-layer edits for each sample. They find that golden layers located on a smaller proxy dataset still produce strong results on new queries drawn from separate test collections. To locate the layers without repeated full edits, the work introduces Layer Gradient Analysis that scores layers through gradient attribution.

Core claim

Fixed golden layers exist that achieve near-optimal editing performance similar to sample-wise optimal layers. These golden layers can be reliably identified using a proxy dataset and generalize effectively to unseen test set queries across datasets. Layer Gradient Analysis estimates golden layers efficiently via gradient-attribution, avoiding extensive trial-and-error across multiple editing runs. Experiments on benchmark datasets confirm effectiveness and robustness across LLM types and knowledge editing methods.

What carries the argument

Layer Gradient Analysis (LGA), which scores layers by gradient attribution to locate golden layers without running full edits on every candidate.

If this is right

  • A single fixed layer can be used for all edits instead of searching per query.
  • Layer selection cost drops because gradient scoring replaces repeated editing trials.
  • Golden layers found on proxy data transfer to new queries and datasets.
  • The same fixed-layer approach works with multiple editing methods and model families.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Golden layers may mark depths where factual knowledge sits in a form that is both accessible and stable.
  • Once identified, the same layer could support repeated fact updates in a deployed model with low overhead.
  • The gradient method might be tested on other model changes such as style or safety adjustments.
  • If golden layers turn out stable across model scales, editing pipelines could standardize on one pre-chosen layer.

Load-bearing premise

A golden layer identified once on a proxy dataset will keep delivering near-optimal edits on new queries without needing to be re-selected for each fresh set of edits.

What would settle it

On a large held-out test collection, the average editing success rate using the fixed golden layer falls well below the average success rate obtained by picking the individually best layer for each query.

Figures

Figures reproduced from arXiv: 2602.20207 by Anshuman Chhabra, Hongfu Liu, Shrestha Datta.

Figure 1
Figure 1. Figure 1: Performance of golden layers selected via the proxy and test sets with GPT-2 XL on (A) [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of model layers for GPT-2 XL, LLaMA2-7B, and Gemma3-12B, where each cell indicates the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance comparison between LGA and CMA across different LLMs and the (A) [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Analyzing the runtime of LGA and CMA over layer-wise [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance of golden layers selected via the proxy and test sets with GPT-2 XL on (A) [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
read the original abstract

Knowledge editing in Large Language Models (LLMs) aims to update the model's prediction for a specific query to a desired target while preserving its behavior on all other inputs. This process typically involves two stages: identifying the layer to edit and performing the parameter update. Intuitively, different queries may localize knowledge at different depths of the model, resulting in different sample-wise editing performance for a fixed editing layer. In this work, we hypothesize the existence of fixed golden layers that can achieve near-optimal editing performance similar to sample-wise optimal layers. To validate this hypothesis, we provide empirical evidence by comparing golden layers against ground-truth sample-wise optimal layers. Furthermore, we show that golden layers can be reliably identified using a proxy dataset and generalize effectively to unseen test set queries across datasets. Finally, we propose a novel method, namely Layer Gradient Analysis (LGA) that estimates golden layers efficiently via gradient-attribution, avoiding extensive trial-and-error across multiple editing runs. Extensive experiments on several benchmark datasets demonstrate the effectiveness and robustness of our LGA approach across different LLM types and various knowledge editing methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper hypothesizes the existence of fixed 'golden layers' in LLMs that achieve near-optimal knowledge editing performance comparable to per-sample optimal layers. It validates this via empirical comparisons to ground-truth sample-wise optima, demonstrates that such layers can be identified from a proxy dataset and generalize across datasets to unseen queries, and introduces Layer Gradient Analysis (LGA) to locate them efficiently using gradient attribution rather than exhaustive editing trials. Extensive experiments on benchmark datasets are reported to show effectiveness and robustness across LLMs and editing methods.

Significance. If the results hold, the work offers a practical advance in knowledge editing by replacing sample-wise layer search with a fixed, efficiently computable layer choice, reducing computational overhead while preserving performance. The direct comparison against sample-wise optima and the proxy-to-test generalization experiments constitute clear strengths; the gradient-based LGA method is a further positive contribution if it reliably ranks layers without multiple full edits.

major comments (2)
  1. [Abstract] Abstract: the central generalization claim—that proxy-identified golden layers transfer reliably to unseen test queries—rests on the unexamined assumption that knowledge-localization patterns are stable across the proxy and test distributions; the manuscript provides no explicit controls for distribution shift (e.g., domain or query-complexity mismatch) or quantification of the performance gap relative to sample-wise optima, which directly affects whether the fixed-layer hypothesis is practically useful.
  2. [Experiments] Experimental results (as summarized in the abstract): no error bars, statistical significance tests, or variance estimates are reported for the claimed improvements of LGA over baselines, making it impossible to assess whether observed gains over sample-wise or other methods are robust or merely within noise.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'near-optimal editing performance similar to sample-wise optimal layers' is used without a quantitative threshold or distance metric; defining 'near-optimal' (e.g., within X% of sample-wise success rate) would sharpen the hypothesis.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address the two major comments point by point below. We agree that additional analysis on distribution shift and statistical reporting will strengthen the paper and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central generalization claim—that proxy-identified golden layers transfer reliably to unseen test queries—rests on the unexamined assumption that knowledge-localization patterns are stable across the proxy and test distributions; the manuscript provides no explicit controls for distribution shift (e.g., domain or query-complexity mismatch) or quantification of the performance gap relative to sample-wise optima, which directly affects whether the fixed-layer hypothesis is practically useful.

    Authors: We appreciate this observation. Our current experiments demonstrate generalization by identifying golden layers on proxy datasets and evaluating on held-out test queries across multiple distinct benchmark datasets, which provides some evidence of stability. However, we agree that explicit controls for distribution shift (such as domain or complexity mismatches) and direct quantification of the performance gap to sample-wise optima are not sufficiently highlighted. In the revision, we will add a new subsection with controlled experiments varying proxy-test distribution differences and will report average, median, and worst-case performance gaps (in terms of editing success rate and perplexity) relative to per-sample optima across all settings. revision: yes

  2. Referee: [Experiments] Experimental results (as summarized in the abstract): no error bars, statistical significance tests, or variance estimates are reported for the claimed improvements of LGA over baselines, making it impossible to assess whether observed gains over sample-wise or other methods are robust or merely within noise.

    Authors: We acknowledge this shortcoming in the presentation of results. While the experiments were conducted over multiple random seeds and initializations, variance information was omitted from the main tables and figures. In the revised manuscript, we will include error bars (standard deviation across runs) in all performance tables and plots. We will also add statistical significance tests (paired t-tests with p-values) comparing LGA against the sample-wise baseline and other methods to confirm that reported improvements are not due to noise. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical validation is self-contained

full rationale

The paper's core claim—that fixed golden layers exist and can be identified via proxy dataset to generalize to test queries—is advanced through direct empirical comparison against sample-wise optimal layers and cross-dataset performance metrics. The LGA method estimates layers using gradient-attribution without any reduction to fitted parameters renamed as predictions, self-definitional equations, or load-bearing self-citations. All steps rely on observable editing success rates rather than constructional equivalence to inputs, rendering the derivation independent and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work is primarily empirical and does not introduce new mathematical axioms or invented physical entities. Standard assumptions of gradient-based attribution in neural networks are used without explicit enumeration.

pith-pipeline@v0.9.0 · 5729 in / 1071 out tokens · 31805 ms · 2026-05-21T11:36:41.941180+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 2 internal anchors

  1. [1]

    Overcoming Catastrophic Forgetting in Neural Networks.Proceedings of the National Academy of Sciences, 2017

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. Overcoming Catastrophic Forgetting in Neural Networks.Proceedings of the National Academy of Sciences, 2017

    work page 2017

  2. [2]

    An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-Tuning.IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2025

    Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, and Yue Zhang. An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-Tuning.IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2025

    work page 2025

  3. [3]

    Editing Large Language Models: Problems, Methods, and Opportunities

    Yunzhi Yao, Peng Wang, Bozhong Tian, Siyuan Cheng, Zhoubo Li, Shumin Deng, Huajun Chen, and Ningyu Zhang. Editing Large Language Models: Problems, Methods, and Opportunities. InEmpirical Methods in Natural Language Processing, 2023

    work page 2023

  4. [4]

    EasyEdit: An Easy-to-Use Knowledge Editing Framework for Large Language Models

    Peng Wang, Ningyu Zhang, Bozhong Tian, Zekun Xi, Yunzhi Yao, Ziwen Xu, Mengru Wang, Shengyu Mao, Xiaohan Wang, Siyuan Cheng, et al. EasyEdit: An Easy-to-Use Knowledge Editing Framework for Large Language Models. InAssociation for Computational Linguistics, 2024

    work page 2024

  5. [5]

    Understanding the Side Effects of Rank-One Knowledge Editing

    Ryosuke Takahashi, Go Kamoda, Benjamin Heinzerling, Keisuke Sakaguchi, and Kentaro Inui. Understanding the Side Effects of Rank-One Knowledge Editing. InBlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, 2025

    work page 2025

  6. [6]

    Does Localization Inform Editing? Surprising Differences in Causality-based Localization vs

    Peter Hase, Mohit Bansal, Been Kim, and Asma Ghandeharioun. Does Localization Inform Editing? Surprising Differences in Causality-based Localization vs. Knowledge Editing in Language Models. InAdvances in Neural Information Processing Systems, 2023

    work page 2023

  7. [7]

    Locating and Editing Factual Associations in GPT

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. Locating and Editing Factual Associations in GPT. InAdvances in Neural Information Processing Systems, 2022

    work page 2022

  8. [8]

    PhD thesis, The University of North Carolina at Chapel Hill, 2024

    Peter Hase.Interpretable and Controllable Language Models. PhD thesis, The University of North Carolina at Chapel Hill, 2024

    work page 2024

  9. [9]

    On the Feasibility of In-Context Probing for Data Attribution

    Cathy Jiao, Weizhen Gao, Aditi Raghunathan, and Chenyan Xiong. On the Feasibility of In-Context Probing for Data Attribution. InFindings of the North American Chapter of the Association for Computational Linguistics, 2025

    work page 2025

  10. [10]

    Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models

    Anshuman Chhabra, Bo Li, Jian Chen, Prasant Mohapatra, and Hongfu Liu. Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models. InInternational Conference on Machine Learning, 2025

    work page 2025

  11. [11]

    Estimating Training Data Influence by Tracing Gradient Descent

    Garima Pruthi, Frederick Liu, Satyen Kale, and Mukund Sundararajan. Estimating Training Data Influence by Tracing Gradient Descent. InAdvances in Neural Information Processing Systems, 2020

    work page 2020

  12. [12]

    Rebuilding ROME: Resolving Model Collapse during Sequential Model Editing

    Akshat Gupta, Sidharth Baskaran, and Gopala Anumanchipalli. Rebuilding ROME: Resolving Model Collapse during Sequential Model Editing. InEmpirical Methods in Natural Language Processing, 2024

    work page 2024

  13. [13]

    Mass-Editing Memory in a Transformer

    Kevin Meng, Arnab Sen Sharma, Alex J Andonian, Yonatan Belinkov, and David Bau. Mass-Editing Memory in a Transformer. InInternational Conference on Learning Representations, 2023. 9 Golden Layers and Where to Find Them

    work page 2023

  14. [14]

    A Unified Framework for Model Editing

    Akshat Gupta, Dev Sajnani, and Gopala Anumanchipalli. A Unified Framework for Model Editing. InFindings of the Empirical Methods in Natural Language Processing, 2024

    work page 2024

  15. [15]

    PMET: Precise Model Editing in a Transformer

    Xiaopeng Li, Shasha Li, Shezheng Song, Jing Yang, Jun Ma, and Jie Yu. PMET: Precise Model Editing in a Transformer. InAAAI Conference on Artificial Intelligence, 2024

    work page 2024

  16. [16]

    Direct and Indirect Effects

    Judea Pearl. Direct and Indirect Effects. InProbabilistic and causal inference: the works of Judea Pearl. Association for Computing Machinery, 2022

    work page 2022

  17. [17]

    Investigating Gender Bias in Language Models Using Causal Mediation Analysis

    Jesse Vig, Sebastian Gehrmann, Yonatan Belinkov, Sharon Qian, Daniel Nevo, Yaron Singer, and Stuart Shieber. Investigating Gender Bias in Language Models Using Causal Mediation Analysis. InAdvances in Neural Information Processing Systems, 2020

    work page 2020

  18. [18]

    Represen- tation Shattering in Transformers: A Synthetic Study with Knowledge Editing.arXiv preprint arXiv:2410.17194, 2024

    Kento Nishi, Rahul Ramesh, Maya Okawa, Mikail Khona, Hidenori Tanaka, and Ekdeep Singh Lubana. Represen- tation Shattering in Transformers: A Synthetic Study with Knowledge Editing.arXiv preprint arXiv:2410.17194, 2024

    work page arXiv 2024

  19. [19]

    Sanity Checks for Saliency Maps

    Julius Adebayo, Justin Gilmer, Michael Muelly, Ian Goodfellow, Moritz Hardt, and Been Kim. Sanity Checks for Saliency Maps. InAdvances in Neural Information Processing Systems, 2018

    work page 2018

  20. [20]

    Axiomatic Attribution for Deep Networks

    Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic Attribution for Deep Networks. InInternational Conference on Machine Learning, 2017

    work page 2017

  21. [21]

    Understanding Black-Box Predictions via Influence Functions

    Pang Wei Koh and Percy Liang. Understanding Black-Box Predictions via Influence Functions. InInternational Conference on Machine Learning, 2017

    work page 2017

  22. [22]

    Revisit, extend, and enhance hessian-free influence functions.CoRR, abs/2405.17490, 2024

    Ziao Yang, Han Yue, Jian Chen, and Hongfu Liu. Revisit, Extend, and Enhance Hessian-Free Influence Functions. arXiv preprint arXiv:2405.17490, 2024

    work page arXiv 2024

  23. [23]

    What Data Benefits My Classifier? Enhancing Model Performance and Interpretability through Influence-Based Data Selection

    Anshuman Chhabra, Peizhao Li, Prasant Mohapatra, and Hongfu Liu. What Data Benefits My Classifier? Enhancing Model Performance and Interpretability through Influence-Based Data Selection. InInternational Conference on Learning Representations, 2024

    work page 2024

  24. [24]

    Make Every Example Count: On the Stability and Utility of Self-Influence for Learning from Noisy NLP Datasets

    Irina Bejan, Artem Sokolov, and Katja Filippova. Make Every Example Count: On the Stability and Utility of Self-Influence for Learning from Noisy NLP Datasets. InEmpirical Methods in Natural Language Processing, 2023

    work page 2023

  25. [25]

    LayerIF: Estimating Layer Quality for Large Language Models using Influence Functions

    Hadi Askari, Shivanshu Gupta, Fei Wang, Anshuman Chhabra, and Muhao Chen. LayerIF: Estimating Layer Quality for Large Language Models using Influence Functions. InAdvances in Neural Information Processing Systems, 2025

    work page 2025

  26. [26]

    First is Not Really Better Than Last: Evaluating Layer Choice and Aggregation Strategies in Language Model Data Influence Estimation

    Dmytro Vitel and Anshuman Chhabra. First is Not Really Better Than Last: Evaluating Layer Choice and Aggregation Strategies in Language Model Data Influence Estimation. InInternational Conference on Learning Representations, 2026

    work page 2026

  27. [27]

    First is Better than Last for Language Data Influence

    Chih-Kuan Yeh, Ankur Taly, Mukund Sundararajan, Frederick Liu, and Pradeep Ravikumar. First is Better than Last for Language Data Influence. InAdvances in Neural Information Processing Systems, 2022

    work page 2022

  28. [28]

    Transformer Feed-Forward Layers are Key-Value Memories

    Mor Geva, Roei Schuster, Jonathan Berant, and Omer Levy. Transformer Feed-Forward Layers are Key-Value Memories. InEmpirical Methods in Natural Language Processing, 2021

    work page 2021

  29. [29]

    Dissecting Recall of Factual Associations in Auto-Regressive Language Models

    Mor Geva, Jasmijn Bastings, Katja Filippova, and Amir Globerson. Dissecting Recall of Factual Associations in Auto-Regressive Language Models. InEmpirical Methods in Natural Language Processing, 2023

    work page 2023

  30. [30]

    Transformer Feed-Forward Layers build Predictions by Promoting Concepts in the Vocabulary Space

    Mor Geva, Avi Caciularu, Kevin Wang, and Yoav Goldberg. Transformer Feed-Forward Layers build Predictions by Promoting Concepts in the Vocabulary Space. InEmpirical Methods in Natural Language Processing, 2022

    work page 2022

  31. [31]

    Shortgpt: Layers in Large Language Models are More Redundant than You Expect

    Xin Men, Mingyu Xu, Qingyu Zhang, Qianhao Yuan, Bingning Wang, Hongyu Lin, Yaojie Lu, Xianpei Han, and Weipeng Chen. Shortgpt: Layers in Large Language Models are More Redundant than You Expect. InFindings of the Association for Computational Linguistics, 2025

    work page 2025

  32. [32]

    How Large Language Models Encode Context Knowledge? A Layer-Wise Probing Study

    Tianjie Ju, Weiwei Sun, Wei Du, Xinwei Yuan, Zhaochun Ren, and Gongshen Liu. How Large Language Models Encode Context Knowledge? A Layer-Wise Probing Study. InJoint International Conference on Computational Linguistics, Language Resources and Evaluation, 2024

    work page 2024

  33. [33]

    Does Knowledge Localization Hold True? Surprising Differences between Entity and Relation Perspectives in Language Models

    Yifan Wei, Xiaoyan Yu, Yixuan Weng, Huanhuan Ma, Yuanzhe Zhang, Jun Zhao, and Kang Liu. Does Knowledge Localization Hold True? Surprising Differences between Entity and Relation Perspectives in Language Models. InConference on Information and Knowledge Management, 2024

    work page 2024

  34. [34]

    What mat- ters in transformers? not all attention is needed.arXiv preprint arXiv:2406.15786, 2024

    Shwai He, Guoheng Sun, Zheyu Shen, and Ang Li. What Matters in Transformers? Not All Attention is Needed. arXiv preprint arXiv:2406.15786, 2024. 10 Golden Layers and Where to Find Them

    work page arXiv 2024

  35. [35]

    Language Models are Unsupervised Multitask Learners.OpenAI blog, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language Models are Unsupervised Multitask Learners.OpenAI blog, 2019

    work page 2019

  36. [36]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open Foundation and Fine-Tuned Chat Models. arXiv preprint arXiv:2307.09288, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  37. [37]

    Gemma 3 Technical Report

    Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, et al. Gemma 3 Technical Report.arXiv preprint arXiv:2503.19786, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    Zero-Shot Relation Extraction via Reading Comprehension

    Omer Levy, Minjoon Seo, Eunsol Choi, and Luke Zettlemoyer. Zero-Shot Relation Extraction via Reading Comprehension. InConference on Computational Natural Language Learning, 2017

    work page 2017

  39. [39]

    Aging with Grace: Lifelong Model Editing with Discrete K-Value Adaptors

    Tom Hartvigsen, Swami Sankaranarayanan, Hamid Palangi, Yoon Kim, and Marzyeh Ghassemi. Aging with Grace: Lifelong Model Editing with Discrete K-Value Adaptors. InAdvances in Neural Information Processing Systems, 2023

    work page 2023

  40. [40]

    A Comprehensive Study of Knowledge Editing for Large Language Models

    Ningyu Zhang, Yunzhi Yao, Bozhong Tian, Peng Wang, Shumin Deng, Mengru Wang, Zekun Xi, Shengyu Mao, Jintian Zhang, Yuansheng Ni, et al. A Comprehensive Study of Knowledge Editing for Large Language Models. arXiv preprint arXiv:2401.01286, 2024

    work page arXiv 2024

  41. [41]

    Evaluating the Ripple Effects of Knowledge Editing in Language Models.Transactions of the Association for Computational Linguistics, 2024

    Roi Cohen, Eden Biran, Ori Yoran, Amir Globerson, and Mor Geva. Evaluating the Ripple Effects of Knowledge Editing in Language Models.Transactions of the Association for Computational Linguistics, 2024

    work page 2024

  42. [42]

    Performance of Some Resistant Rules for Outlier Labeling

    David C Hoaglin, Boris Iglewicz, and John W Tukey. Performance of Some Resistant Rules for Outlier Labeling. Journal of the American Statistical Association, 1986. 11 Golden Layers and Where to Find Them Appendix A Implementation Details Here we introduce the implementation details for knowledge editing in terms of datasets, models, editing methods, and e...

    work page arXiv 1986