mlsanityData to 14 Oct 2026

FALCON: Fine-grained Activation Manipulation by Contrastive Orthogonal Unalignment for Large Language Model

arXiv 2502.01472 ↗First posted 3 Feb 2025cs.CL cs.AIcited by 1

On the map: LLM alignment / RLHF / preference optimisation → LLM preference optimization methods

also in 1 more
Abstract (arXiv)

Large language models have been widely applied, but can inadvertently encode sensitive or harmful information, raising significant safety concerns. Machine unlearning has emerged to alleviate this concern; however, existing training-time unlearning approaches, relying on coarse-grained loss combinations, have limitations in precisely separating knowledge and balancing removal effectiveness with model utility. In contrast, we propose Fine-grained Activation manipuLation by Contrastive Orthogonal uNalignment (FALCON), a novel representation-guided unlearning approach that leverages information-theoretic guidance for efficient parameter selection, employs contrastive mechanisms to enhance representation separation, and projects conflict gradients onto orthogonal subspaces to resolve conflicts between forgetting and retention objectives. Extensive experiments demonstrate that FALCON achieves superior unlearning effectiveness while maintaining model utility, exhibiting robust resistance against knowledge recovery attempts.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 69 of its 95 references that are in the map. First posted on arXiv 3 Feb 2025.

Builds on

69 of 95 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning9 Oct
  2. 2024Spectral Editing of Activations for Large Language Model Alignment15 May
  3. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
  4. 2022Training language models to follow instructions with human feedback4 Mar

Other references · 65

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  5. 2023Mistral 7Bcited by 3,164
  6. 2016Pointer Sentinel Mixture Modelscited by 3,096
  7. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  8. 2022Editing Models with Task Arithmeticcited by 1,120
  9. 2024Yi: Open Foundation Models by 01.AIcited by 719
  10. 2018Predictive Uncertainty Estimation via Prior Networkscited by 557
  11. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 533
  12. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
38 more
  1. 2020BERT-ATTACK: Adversarial Attack Against BERT Using BERTcited by 491
  2. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Modelscited by 437
  3. 2023Who's Harry Potter? Approximate Unlearning in LLMscited by 381
  4. 2024Improving Alignment and Robustness with Circuit Breakerscited by 322
  5. 2019On Mutual Information Maximization for Representation Learningcited by 316
  6. 2023Large Language Model Unlearningcited by 302
  7. 2021Unsolved Problems in ML Safetycited by 292
  8. 2023Can We Edit Factual Knowledge by In-Context Learning?cited by 285
  9. 2023Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-trainingcited by 265
  10. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Modelscited by 261
  11. 2024Rethinking Machine Unlearning for Large Language Modelscited by 257
  12. 2023In-Context Unlearning: Language Models as Few Shot Unlearnerscited by 224
  13. 2023Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampeningcited by 181
  14. 2024A Comprehensive Study of Knowledge Editing for Large Language Modelscited by 163
  15. 2023Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attackscited by 158
  16. 2023Composing Parameter-Efficient Modules with Arithmetic Operationscited by 156
  17. 2021Self-Supervised Representation Learning: Introduction, Advances and Challengescited by 148
  18. 2022Domain Generalization by Mutual-Information Regularization with Pre-trained Modelscited by 133
  19. 2024Safe LoRA: the Silver Lining of Reducing Safety Risks when Fine-tuning Large Language Modelscited by 125
  20. 2024An Adversarial Perspective on Machine Unlearning for AI Safetycited by 123
  21. 2024Large Language Model Unlearning via Embedding-Corrupted Promptscited by 116
  22. 2024SOUL: Unlocking the Power of Second-Order Optimization for LLM Unlearningcited by 104
  23. 2024Representation Noising: A Defence Mechanism Against Harmful Finetuningcited by 96
  24. 2024Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Differencecited by 88
  25. 2024Safeguarding Large Language Models: A Surveycited by 78
  26. 2024Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradientcited by 69
  27. 2023Learning to Unlearn: Instance-wise Unlearning for Pre-trained Classifierscited by 67
  28. 2023Right to be Forgotten in the Era of Large Language Models: Implications, Challenges, and Solutionscited by 60
  29. 2022Entropy-based Attention Regularization Frees Unintended Bias Mitigation from Listscited by 55
  30. 2018Entropy and mutual information in models of deep neural networkscited by 51
  31. 2023Exploring the Landscape of Machine Unlearning: A Comprehensive Survey and Taxonomycited by 51
  32. 2024UNDIAL: Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Modelscited by 50
  33. 2024Offset Unlearning for Large Language Modelscited by 36
  34. 2022Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlinear Guarded Attribute Informationcited by 34
  35. 2024ROSE Doesn't Do That: Boosting the Safety of Instruction-Tuned Large Language Models with Reverse Prompt Contrastive Decodingcited by 32
  36. 2023InfoPrompt: Information-Theoretic Soft Prompt Tuning for Natural Language Understandingcited by 30
  37. 2024Navigating LLM Ethics: Advancements, Challenges, and Future Directionscited by 28
  38. 2024WAGLE: Strategic Weight Attribution for Effective and Modular Unlearning in Large Language Modelscited by 26

50 most cited of 65 All 65 (JSON)

Built on by

Cited by year

Citing papers per year: 2025: 1.

1 paper.

  1. 2025SoK: Machine Unlearning for Large Language Modelscited by 11

Similar, not cited

Close in wording, no citation either way

  1. 2025UniErase: Towards Balanced and Precise Unlearning in Language Models
  2. 2026Inference-Time Machine Unlearning via Gated Activation Redirection
  3. 2025UCD: Unlearning in LLMs via Contrastive Decoding
  4. 2026CAP: Controllable Alignment Prompting for Unlearning in LLMs
  5. 2026ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models
  6. 2025Model Unlearning via Sparse Autoencoder Subspace Guided Projections
  7. 2026SAUL: Sharpness-Aware Augmented-Lagrangian Unlearning
  8. 2026Null-Space Constrained Low-Rank Adaptation for Response-Specified Large Language Model Unlearning