pith. machine review for the scientific record. sign in

arxiv: 2605.10680 · v1 · submitted 2026-05-11 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

Exact Unlearning from Proxies Induces Closeness Guarantees on Approximate Unlearning

Virgile Dine , Teddy Furon
Authors on Pith no claims yet

Pith reviewed 2026-05-12 03:25 UTC · model grok-4.3

classification 💻 cs.LG
keywords machine unlearningexact unlearningapproximate unlearningKL divergencedata distributionsproxy modelsadmissibility criterionforgetting scenarios
0
0 comments X

The pith

Linking machine unlearning to data distributions via proxies provides KL divergence bounds to ideal retrained models.

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

The paper proposes a new approach to machine unlearning that focuses on the structure of data distributions instead of directly updating neural network parameters. By inferring these distributions precisely, it becomes possible to distill an exact unlearning signal. This results in theoretical bounds on the Kullback-Leibler divergence between the unlearned model and the ideal model from retraining from scratch, under a verifiable admissibility criterion. Validation through experiments on three forgetting scenarios shows this method yields the closest classifier to the ideal retrained model among compared techniques.

Core claim

Exact unlearning from proxies of the data distributions induces closeness guarantees on approximate unlearning, specifically theoretical bounds on the Kullback-Leibler divergence from the ideal retrained model to the unlearned model, under a verifiable admissibility criterion.

What carries the argument

Exact unlearning from proxies, which links unlearning directly to data distribution structure to distill the unlearning signal and provide closeness guarantees.

If this is right

  • The approximate unlearned model is bounded in its divergence from the ideal retrained model.
  • The framework is shown to be sound through these theoretical bounds.
  • The method achieves the closest performance to ideal retraining in experimental forgetting scenarios.
  • The admissibility criterion allows practical verification of the guarantees.

Where Pith is reading between the lines

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

  • This suggests that accurate distribution modeling could make unlearning more reliable without always requiring full retraining.
  • Extending this to non-classification tasks or other models could broaden the applicability of guaranteed unlearning.
  • Challenges in precisely inferring distributions might limit adoption, pointing to needs for robust inference techniques.

Load-bearing premise

Inferring the data distributions with enough precision to distill the exact unlearning signal, and that the admissibility criterion is verifiable in practice.

What would settle it

A case where the Kullback-Leibler divergence between the unlearned model and the retrained model exceeds the paper's theoretical bound, even though the admissibility criterion is satisfied.

Figures

Figures reproduced from arXiv: 2605.10680 by Teddy Furon, Virgile Dine.

Figure 2
Figure 2. Figure 2: 6 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Convexity of h(η) on DINOv2+mlp1/CIFAR-10 for 3 forgetting scenarios and 5 proxy models: LDA (9), QDA (10) (dashed), LDA-Mix (14), QDA-Mix (15) (dashed), LDA-2C (16). 7 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) KLt as a function of RTE normalised by the Retrain RTE on CIFAR-10 / DINOv2 + MLP-1, scenario Subclass (airplane). (b) Number of queries an attacker would need to distinguish the unlearned model from the retrained model up to an α rate of false positive. 9 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Cross-entropy training matches LDA. Top: last-layer weight matrix (rows reshaped as images) of a linear network trained with the Adam optimizer and cross-entropy loss. Bottom: closed-form LDA weight formula. Dataset: MNIST [LeCun et al., 1998]. The near-perfect visual match empirically confirms the cross-entropy/LDA equivalence. Cost of Homoscedastic Assumption This subsection exhibits the cost to model th… view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the sub-class scenario over synthetic heteroscedastic Gaussian data. Class 1 [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
read the original abstract

This paper proposes a paradigm shift linking machine unlearning directly to the structure of the data distributions rather than a mere update of the neural network parameters. We show that inferring these distributions with precision enables distilling the exact unlearning signal induced by the modeling. Theoretical bounds on the Kullback-Leibler divergence from the ideal retrained model to our unlearned model, under verifiable admissibility criterion, reveal the soundness of our framework. This method is experimentally validated over three forgetting scenarios as reaching the closest classifier to the ideal retrained model when compared to competitors.

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 / 2 minor

Summary. The paper proposes a paradigm shift in machine unlearning by connecting it directly to the structure of data distributions instead of parameter updates. It claims that precise inference of distributions allows distilling an exact unlearning signal, derives theoretical bounds on the KL divergence from the ideal retrained model to the unlearned model under a verifiable admissibility criterion, and experimentally shows the method produces the closest classifier to the retrained model across three forgetting scenarios relative to competitors.

Significance. If the KL bounds are rigorously established and the experiments demonstrate clear superiority with proper controls, the work would be significant for providing distribution-based guarantees in unlearning, a key open challenge. This could shift the field from heuristic updates toward verifiable closeness to retraining, with potential impact on privacy and compliance applications.

major comments (2)
  1. [Abstract] Abstract: the central claim of theoretical KL bounds under a 'verifiable admissibility criterion' is asserted without any derivation outline, definition of the criterion, or statement of assumptions. This is load-bearing for the soundness argument and cannot be assessed from the given information.
  2. [Abstract] Abstract and experimental section: the claim of reaching the 'closest classifier' is presented without error bars, statistical significance tests, or details on how closeness is measured across the three forgetting scenarios. This undermines the experimental validation of superiority.
minor comments (2)
  1. The weakest assumption (precise distribution inference enabling signal distillation) should be discussed with concrete examples or failure modes to clarify practical verifiability.
  2. Notation for the unlearned model, proxy, and ideal retrained model should be introduced early and used consistently to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and will incorporate revisions to strengthen the presentation of our theoretical and experimental contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of theoretical KL bounds under a 'verifiable admissibility criterion' is asserted without any derivation outline, definition of the criterion, or statement of assumptions. This is load-bearing for the soundness argument and cannot be assessed from the given information.

    Authors: We agree that the abstract, as a high-level summary, should provide sufficient context for the central theoretical claim. The full manuscript contains the complete derivation of the KL bounds, the definition of the verifiable admissibility criterion, and the explicit list of assumptions in the theoretical development. In the revised version, we will expand the abstract to include a concise outline of the derivation steps, a brief definition of the admissibility criterion, and the key assumptions, ensuring the claim is assessable without requiring the reader to consult the body of the paper. revision: yes

  2. Referee: [Abstract] Abstract and experimental section: the claim of reaching the 'closest classifier' is presented without error bars, statistical significance tests, or details on how closeness is measured across the three forgetting scenarios. This undermines the experimental validation of superiority.

    Authors: We acknowledge that additional statistical details would strengthen the experimental claims. The manuscript reports closeness via KL divergence to the retrained model across the three scenarios, with comparisons to baselines. In the revision, we will add error bars to all reported metrics, include statistical significance tests (such as paired t-tests with p-values), and explicitly detail the measurement procedure for closeness in both the abstract and the experimental section to provide rigorous validation of the superiority results. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and context describe a framework that infers data distributions to distill an unlearning signal, then derives KL divergence bounds to the retrained model under an admissibility criterion, with experimental validation across scenarios. No load-bearing derivation steps, equations, or self-citations are exhibited in the available text that reduce by construction to fitted inputs, self-definitions, or prior author work. The central claims rest on independent theoretical bounds and direct empirical comparisons rather than renaming or smuggling ansatzes. This is the expected honest non-finding for a paper whose derivation chain is not shown to collapse internally.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.0 · 5385 in / 996 out tokens · 25600 ms · 2026-05-12T03:25:01.165005+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

107 extracted references · 107 canonical work pages · 1 internal anchor

  1. [1]

    and Thomas, J.A

    Cover, T.M. and Thomas, J.A. , biburl =

    work page

  2. [2]

    2015 IEEE Symposium on Security and Privacy , pages =

    Towards Making Systems Forget with Machine Unlearning , author =. 2015 IEEE Symposium on Security and Privacy , pages =. 2015 , organization =

    work page 2015

  3. [3]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    work page

  4. [4]

    Proceedings of the 37th International Conference on Machine Learning (ICML) , year =

    Certified Data Removal from Machine Learning Models , author =. Proceedings of the 37th International Conference on Machine Learning (ICML) , year =

    work page

  5. [5]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Amnesiac Machine Learning , author =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

    work page

  6. [6]

    2017 IEEE Symposium on Security and Privacy (SP) , year =

    Membership Inference Attacks Against Machine Learning Models , author =. 2017 IEEE Symposium on Security and Privacy (SP) , year =

    work page 2017

  7. [7]

    28th USENIX Security Symposium (USENIX Security 19) , year =

    The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks , author =. 28th USENIX Security Symposium (USENIX Security 19) , year =

    work page

  8. [8]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , title =

    Chen, Min and Gao, Weizhuo and Liu, Gaoyang and Peng, Kai and Wang, Chen , year =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , title =

    work page

  9. [9]

    Transactions on Machine Learning Research , year =

    Oquab, Maxime and Darcet, Timoth. Transactions on Machine Learning Research , year =

    work page

  10. [10]

    2016 , institution =

    Regulation (EU) 2016/679 (General Data Protection Regulation) , author =. 2016 , institution =

    work page 2016

  11. [11]

    , title =

    Csiszár, I. , title =. The Annals of Probability , year =

    work page

  12. [12]

    Choquette-Choo and Hengrui Jia and Adelin Travers and Baiwu Zhang and David Lie and Nicolas Papernot , title =

    Lucas Bourtoule and Varun Chandrasekaran and Christopher A. Choquette-Choo and Hengrui Jia and Adelin Travers and Baiwu Zhang and David Lie and Nicolas Papernot , title =. 2021. 2021 , publisher =

    work page 2021

  13. [13]

    Proceedings of the 34th International Conference on Machine Learning (

    Pang Wei Koh and Percy Liang , title =. Proceedings of the 34th International Conference on Machine Learning (. 2017 , publisher =

    work page 2017

  14. [14]

    Advances in Neural Information Processing Systems (

    Ayush Sekhari and Jayadev Acharya and Gautam Kamath and Ananda Theertha Suresh , title =. Advances in Neural Information Processing Systems (. 2021 , publisher =

    work page 2021

  15. [15]

    Advances in Neural Information Processing Systems (

    Meghdad Kurmanji and Peter Triantafillou and Jamie Hayes and Eleni Triantafillou , title =. Advances in Neural Information Processing Systems (. 2023 , publisher =

    work page 2023

  16. [16]

    Chundawat and Ayush Kumar Tarun and Murari Mandal and Mohan S

    Vikram S. Chundawat and Ayush Kumar Tarun and Murari Mandal and Mohan S. Kankanhalli , title =. Proceedings of the 37th. 2023 , publisher =

    work page 2023

  17. [17]

    The 12th International Conference on Learning Representations (

    Chongyu Fan and Jiancheng Liu and Yihua Zhang and Eric Wong and Dennis Wei and Sijia Liu , title =. The 12th International Conference on Learning Representations (. 2024 , note =

    work page 2024

  18. [18]

    Chundawat and Murari Mandal and Mohan S

    Ayush Kumar Tarun and Vikram S. Chundawat and Murari Mandal and Mohan S. Kankanhalli , title =. 2023 , publisher =

    work page 2023

  19. [19]

    Lampert , title =

    Alexandra Peste and Dan Alistarh and Christoph H. Lampert , title =. 2021 , note =

    work page 2021

  20. [20]

    2022 , month = oct, journal =

    Thanh Tam Nguyen and Thanh Trung Huynh and Phi Le Nguyen and Alan Wee-Chung Liew and Hongzhi Yin and Quoc Viet Hung Nguyen , title =. arXiv preprint arXiv:2209.02299 , year =

    work page arXiv

  21. [21]

    Reza Shokri and Marco Stronati and Congzheng Song and Vitaly Shmatikov , title =. 2017. 2017 , publisher =

    work page 2017

  22. [22]

    Membership Inference Attacks From First Principles , booktitle =

    Nicholas Carlini and Steve Chien and Milad Nasr and Shuang Song and Andreas Terzis and Florian Tram. Membership Inference Attacks From First Principles , booktitle =. 2022 , publisher =

    work page 2022

  23. [23]

    Zico Kolter , title =

    Pratyush Maini and Zhili Feng and Avi Schwarzschild and Zachary Chase Lipton and J. Zico Kolter , title =. Conference on Language Modeling (

    work page

  24. [24]

    2024 , eprint=

    Are we making progress in unlearning? Findings from the first NeurIPS unlearning competition , author=. 2024 , eprint=

    work page 2024

  25. [25]

    Advances in Neural Information Processing Systems (

    Erik Daxberger and Agustinus Kristiadi and Alexander Immer and Runa Eschenhagen and Matthias Bauer and Philipp Hennig , title =. Advances in Neural Information Processing Systems (. 2021 , publisher =

    work page 2021

  26. [26]

    The 10th International Conference on Learning Representations (

    Ananya Kumar and Aditi Raghunathan and Robbie Jones and Tengyu Ma and Percy Liang , title =. The 10th International Conference on Learning Representations (. 2022 , note =

    work page 2022

  27. [27]

    2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , title =

    Hayes, Jamie and Shumailov, Ilia and Triantafillou, Eleni and Khalifa, Amr and Papernot, Nicolas , year =. 2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , title =

    work page 2025

  28. [28]

    McLachlan , title =

    Geoffrey J. McLachlan , title =. 1992 , edition =

    work page 1992

  29. [29]

    2026 , eprint=

    Softmax Linear Attention: Reclaiming Global Competition , author=. 2026 , eprint=

    work page 2026

  30. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (

    Aditya Golatkar and Alessandro Achille and Avinash Ravichandran and Marzia Polito and Stefano Soatto , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (. 2021 , publisher =

    work page 2021

  31. [31]

    Understanding intermediate layers using linear classifier probes

    Guillaume Alain and Yoshua Bengio , title =. arXiv preprint arXiv:1610.01644 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    Le , title =

    Simon Kornblith and Jonathon Shlens and Quoc V. Le , title =. Proceedings of the. 2019 , publisher =

    work page 2019

  33. [33]

    Alex Krizhevsky , title =

    work page

  34. [34]

    Advances in Neural Information Processing Systems (

    Kimin Lee and Kibok Lee and Honglak Lee and Jinwoo Shin , title =. Advances in Neural Information Processing Systems (. 2018 , publisher =

    work page 2018

  35. [35]

    Advances in Neural Information Processing Systems (

    Yifei Ming and Ziyang Cai and Jiuxiang Gu and Yiyou Sun and Wei Li and Yixuan Li , title =. Advances in Neural Information Processing Systems (. 2022 , publisher =

    work page 2022

  36. [36]

    Proceedings of the 52nd Annual

    Vitaly Feldman , title =. Proceedings of the 52nd Annual. 2020 , publisher =

    work page 2020

  37. [37]

    Friedman and Richard A

    Leo Breiman and Jerome H. Friedman and Richard A. Olshen and Charles J. Stone , title =

    work page

  38. [38]

    Proceedings of the 41st International Conference on Machine Learning (

    Eli Chien and Haoyu Peter Wang and Ziang Chen and Pan Li , title =. Proceedings of the 41st International Conference on Machine Learning (. 2024 , publisher =

    work page 2024

  39. [39]

    Proceedings of the 41st International Conference on Machine Learning (

    Martin Pawelczyk and Seth Neel and Himabindu Lakkaraju , title =. Proceedings of the 41st International Conference on Machine Learning (. 2024 , publisher =

    work page 2024

  40. [40]

    2014 , series =

    Cynthia Dwork and Aaron Roth , title =. 2014 , series =

    work page 2014

  41. [41]

    Zou , title =

    Antonio Ginart and Melody Guan and Gregory Valiant and James Y. Zou , title =. Advances in Neural Information Processing Systems (. 2019 , publisher =

    work page 2019

  42. [42]

    Bishop , title =

    Christopher M. Bishop , title =

    work page

  43. [43]

    Donoho , title =

    Vardan Papyan and X.\ Y.\ Han and David L. Donoho , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

    work page 2020

  44. [44]

    Gradient-Based Learning Applied to Document Recognition , journal =

    Yann LeCun and L. Gradient-Based Learning Applied to Document Recognition , journal =. 1998 , doi =

    work page 1998

  45. [45]

    2025 , eprint=

    Machine Unlearning via Information Theoretic Regularization , author=. 2025 , eprint=

    work page 2025

  46. [46]

    Electronics and Communications in Japan (Part I: Communications) , volume =

    Amari, Shun-Ichi , title =. Electronics and Communications in Japan (Part I: Communications) , volume =. doi:https://doi.org/10.1002/ecja.4400660602 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/ecja.4400660602 , abstract =

    work page doi:10.1002/ecja.4400660602

  47. [47]

    Proceedings of the 38th

    Jack Foster and Stefan Schoepf and Alexandra Brintrup , title =. Proceedings of the 38th. 2024 , publisher =

    work page 2024

  48. [48]

    2026 , eprint=

    POUR: A Provably Optimal Method for Unlearning Representations via Neural Collapse , author=. 2026 , eprint=

    work page 2026

  49. [49]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Ascent Fails to Forget , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    work page

  50. [50]

    2025 , MONTH = Oct, DOI =

    Dine, Virgile and Furon, Teddy and Faure, Charly , URL =. 2025 , MONTH = Oct, DOI =

    work page 2025

  51. [51]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year =

    Efficient Utility-Preserving Machine Unlearning with Implicit Gradient Surgery , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year =

    work page

  52. [52]

    2024 , eprint=

    Deep Unlearning: Fast and Efficient Gradient-free Approach to Class Forgetting , author=. 2024 , eprint=

    work page 2024

  53. [53]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year =

    Unlearning-Aware Minimization , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year =

    work page

  54. [54]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    On the Misalignment Between Data Learnability and Forgettability in Machine Unlearning , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2026 , month=. doi:10.1609/aaai.v40i29.39657 , abstractNote=

    work page doi:10.1609/aaai.v40i29.39657 2026

  55. [55]

    Proceedings of the 43rd International Conference on Machine Learning (ICML) , year =

    Attribute-to-Delete: Machine Unlearning via Datamodel Matching , author =. Proceedings of the 43rd International Conference on Machine Learning (ICML) , year =

    work page

  56. [56]

    ACM Computing Surveys , volume =

    Machine Unlearning: A Survey , author =. ACM Computing Surveys , volume =. 2023 , doi =

    work page 2023

  57. [57]

    ACM Computing Surveys , year =

    A Survey on Federated Unlearning: Challenges, Methods, and Future Directions , author =. ACM Computing Surveys , year =

    work page

  58. [58]

    Nature Machine Intelligence , year =

    Rethinking Machine Unlearning for Large Language Models , author =. Nature Machine Intelligence , year =

    work page

  59. [59]

    Voigt, Paul and Von dem Bussche, Axel , edition =. The. 2017 , publisher =

    work page 2017

  60. [60]

    and Valiant, Gregory and Zou, James Y

    Ginart, Antonio and Guan, Melody Y. and Valiant, Gregory and Zou, James Y. , booktitle =. Making

    work page

  61. [61]

    Proceedings of the 32nd International Conference on Algorithmic Learning Theory (ALT) , series =

    Descent-to-Delete: Gradient-Based Methods for Machine Unlearning , author =. Proceedings of the 32nd International Conference on Algorithmic Learning Theory (ALT) , series =

    work page

  62. [62]

    Unrolling

    Thudi, Anvith and Deza, Gabriel and Chandrasekaran, Varun and Papernot, Nicolas , booktitle =. Unrolling. 2022 , publisher =

    work page 2022

  63. [63]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Model Sparsity Can Simplify Machine Unlearning , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

    work page

  64. [64]

    Proceedings of the 24th International Conference on Artificial Intelligence and Statistics (AISTATS) , series =

    Approximate Data Deletion from Machine Learning Models , author =. Proceedings of the 24th International Conference on Artificial Intelligence and Statistics (AISTATS) , series =

    work page

  65. [65]

    Proceedings of the 30th Network and Distributed System Security Symposium (NDSS) , year =

    Machine Unlearning of Features and Labels , author =. Proceedings of the 30th Network and Distributed System Security Symposium (NDSS) , year =

    work page

  66. [66]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages =

    Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input--Output Observations , author =. Proceedings of the European Conference on Computer Vision (ECCV) , pages =

    work page

  67. [67]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    Deep Unlearning via Randomized Conditionally Independent Hessians , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    work page

  68. [68]

    Variational

    Nguyen, Quoc Phong and Low, Bryan Kian Hsiang and Jaillet, Patrick , booktitle =. Variational

    work page

  69. [69]

    31st USENIX Security Symposium (USENIX Security 22) , pages =

    On the Necessity of Auditable Algorithmic Definitions for Machine Unlearning , author =. 31st USENIX Security Symposium (USENIX Security 22) , pages =

    work page

  70. [70]

    2018 IEEE 31st Computer Security Foundations Symposium (CSF) , pages =

    Privacy Risk in Machine Learning: Analyzing the Connection to Overfitting , author =. 2018 IEEE 31st Computer Security Foundations Symposium (CSF) , pages =. 2018 , publisher =

    work page 2018

  71. [71]

    arXiv preprint arXiv:2201.06640 , year =

    Towards Adversarial Evaluations for Inexact Machine Unlearning , author =. arXiv preprint arXiv:2201.06640 , year =

    work page arXiv

  72. [72]

    arXiv preprint arXiv:2402.14015 , year =

    Corrective Machine Unlearning , author =. arXiv preprint arXiv:2402.14015 , year =

    work page arXiv

  73. [73]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

    Erasing Concepts from Diffusion Models , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

    work page

  74. [74]

    Eldan, Ronen and Russinovich, Mark , journal =. Who's

    work page

  75. [75]

    and Dombrowski, Ann-Kathrin and Goel, Shashwat and Phan, Long and others , booktitle =

    Li, Nathaniel and Pan, Alexander and Gopal, Anjali and Yue, Summer and Berrios, Daniel and Gatti, Alice and Li, Justin D. and Dombrowski, Ann-Kathrin and Goel, Shashwat and Phan, Long and others , booktitle =. The

    work page

  76. [76]

    2021 , publisher =

    Liu, Gaoyang and Ma, Xiaoqiang and Yang, Yang and Wang, Chen and Liu, Jiangchuan , booktitle =. 2021 , publisher =

    work page 2021

  77. [77]

    IEEE Transactions on Neural Networks and Learning Systems , year =

    Federated Unlearning: A Survey on Methods, Design Guidelines, and Evaluation Metrics , author =. IEEE Transactions on Neural Networks and Learning Systems , year =

    work page

  78. [78]

    and Lindenstrauss, Joram , journal =

    Johnson, William B. and Lindenstrauss, Joram , journal =. Extensions of

    work page

  79. [79]

    Proceedings of the 30th Annual ACM Symposium on Theory of Computing (STOC) , pages =

    Approximate nearest neighbors: towards removing the curse of dimensionality , author =. Proceedings of the 30th Annual ACM Symposium on Theory of Computing (STOC) , pages =

    work page

  80. [80]

    An elementary proof of a theorem of

    Dasgupta, Sanjoy and Gupta, Anupam , journal =. An elementary proof of a theorem of

    work page

Showing first 80 references.