mlsanityData to 14 Oct 2026

On Effects of Steering Latent Representation for Large Language Model Unlearning

arXiv 2408.06223 ↗First posted 12 Aug 2024cs.CL cs.AIcited by 32

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Representation Misdirection for Unlearning (RMU), which steers model representation in the intermediate layer to a target random representation, is an effective method for large language model (LLM) unlearning. Despite its high performance, the underlying cause and explanation remain underexplored. In this paper, we theoretically demonstrate that steering forget representations in the intermediate layer reduces token confidence, causing LLMs to generate wrong or nonsense responses. We investigate how the coefficient influences the alignment of forget-sample representations with the random direction and hint at the optimal coefficient values for effective unlearning across different network layers. We show that RMU unlearned models are robust against adversarial jailbreak attacks. Furthermore, our empirical analysis shows that RMU is less effective when applied to the middle and later layers in LLMs. To resolve this drawback, we propose Adaptive RMU--a simple yet effective alternative method that makes unlearning effective with most layers. Extensive experiments demonstrate that Adaptive RMU significantly improves the unlearning performance compared to prior art while incurring no additional computational cost.

Key facts. Cited by 32 papers in the map (data through 14 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 47 of its 66 references that are in the map. First posted on arXiv 12 Aug 2024.

Builds on

47 of 66 references.

Closest ancestors · 17

Same direction, newest first

  1. 2024Machine Unlearning in Generative AI: A Survey30 Jul
  2. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Models8 Jul
  3. 2024Large Language Model Unlearning via Embedding-Corrupted Prompts12 Jun
  4. 2024SOUL: Unlocking the Power of Second-Order Optimization for LLM Unlearning28 Apr
  5. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
  6. 2024Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy2 Mar
  7. 2024Eight Methods to Evaluate Robust Unlearning in LLMs26 Feb
  8. 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
  9. 2023Certified Minimax Unlearning with Generalization Rates and Deletion Capacity16 Dec
  10. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation19 Oct
  11. 2023Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampening15 Aug
  12. 2023SAFE: Machine Unlearning With Shard Graphs25 Apr
5 more
  1. 2023Learning to Unlearn: Instance-wise Unlearning for Pre-trained Classifiers27 Jan
  2. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Models4 Oct
  3. 2021Unrolling SGD: Understanding Factors Influencing Machine Unlearning27 Sep
  4. 2021Machine Unlearning of Features and Labels26 Aug
  5. 2021Remember What You Want to Forget: Algorithms for Machine Unlearning4 Mar

Other references · 30

  1. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  2. 2016A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networkscited by 2,385
  3. 2017Understanding Black-box Predictions via Influence Functionscited by 1,741
  4. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  5. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  6. 2020Energy-based Out-of-distribution Detectioncited by 1,299
  7. 2019Machine Unlearningcited by 1,061
  8. 2024Yi: Open Foundation Models by 01.AIcited by 719
  9. 2019Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networkscited by 618
  10. 2022Out-of-Distribution Detection with Deep Nearest Neighborscited by 600
  11. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  12. 2023Zephyr: Direct Distillation of LM Alignmentcited by 506
18 more
  1. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 465
  2. 2023GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Ciphercited by 423
  3. 2019Confident Learning: Estimating Uncertainty in Dataset Labelscited by 413
  4. 2023Studying Large Language Model Generalization with Influence Functionscited by 326
  5. 2020Approximate Data Deletion from Machine Learning Modelscited by 255
  6. 2023Scalable and Transferable Black-Box Jailbreaks for Language Models via Persona Modulationcited by 214
  7. 2023Automatically Auditing Large Language Models via Discrete Optimizationcited by 191
  8. 2023Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attackscited by 158
  9. 2022Federated Unlearning: How to Efficiently Erase a Client in FL?cited by 95
  10. 2020Learn to Forget: Machine Unlearning via Neuron Maskingcited by 62
  11. 2024Machine Unlearning for Image-to-Image Generative Modelscited by 47
  12. 2023Heterogeneous Federated Knowledge Graph Embedding Learning and Unlearningcited by 40
  13. 2024Towards Effective and General Graph Unlearning via Mutual Evolutioncited by 29
  14. 2024Federated Unlearning: A Survey on Methods, Design Guidelines, and Evaluation Metricscited by 27
  15. 2024Post-Training Attribute Unlearning in Recommender Systemscited by 24
  16. 2024Towards Efficient and Effective Unlearning of Large Language Models for Recommendationcited by 16
  17. 2023Making Users Indistinguishable: Attribute-wise Unlearning in Recommender Systemscited by 15
  18. 2024Unlink to Unlearn: Simplifying Edge Unlearning in GNNscited by 11

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 21; 2026: 9.

32 papers.

Sort:

By area:

  1. 2024An Adversarial Perspective on Machine Unlearning for AI Safetycited by 123
  2. 2024Position: LLM Unlearning Benchmarks are Weak Measures of Progresscited by 63
  3. 2025LLM Unlearning via Neural Activation Redirectioncited by 41
  4. 2025A Comprehensive Survey of Machine Unlearning Techniques for Large Language Modelscited by 34
  5. 2025Activation Steering with a Feedback Controllercited by 15
  6. 2025A General Framework to Enhance Fine-tuning-based LLM Unlearningcited by 14
  7. 2025SoK: Machine Unlearning for Large Language Modelscited by 11
  8. 2025Improving LLM Unlearning Robustness via Random Perturbationscited by 9
  9. 2025Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedycited by 9
  10. 2025Does Machine Unlearning Truly Remove Knowledge?cited by 7
  11. 2026Attention Smoothing Is All You Need For Unlearningcited by 7
  12. 2025Step-by-Step Reasoning Attack: Revealing 'Erased' Knowledge in Large Language Modelscited by 6
20 more
  1. 2025UniErase: Towards Balanced and Precise Unlearning in Language Modelscited by 5
  2. 2025Curvature-Aware Safety Restoration In LLMs Fine-Tuningcited by 5
  3. 2025The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluationcited by 5
  4. 2025Generative Model Unlearning: A Survey through Target Events, Unlearning Operators, and Evaluation Protocolscited by 4
  5. 2025A Survey on Unlearning in Large Language Modelscited by 4
  6. 2025LLM Unlearning Should Be Form-Independentcited by 3
  7. 2026LLMs Can Unlearn Refusal with Only 1,000 Benign Samplescited by 3
  8. 2025Soft Token Attacks Cannot Reliably Audit Unlearning in Large Language Modelscited by 2
  9. 2025AUVIC: Adversarial Unlearning of Visual Concepts for Multi-modal Large Language Modelscited by 2
  10. 2025Feature-Selective Representation Misdirection for Machine Unlearningcited by 2
  11. 2026Beyond Forgetting: Representation Misdirection Elicits Controllable Side Behaviors and Capabilitiescited by 1
  12. 2025Lifting Data-Tracing Machine Unlearning to Knowledge-Tracing for Foundation Modelscited by 0
  13. 2025Mr. Snuffleupagus at SemEval-2025 Task 4: Unlearning Factual Knowledge from LLMs Using Adaptive RMUcited by 0
  14. 2025RapidUn: Influence-Driven Parameter Reweighting for Efficient Large Language Model Unlearningcited by 0
  15. 2026EvoMU: Evolutionary Machine Unlearningcited by 0
  16. 2026Fuel Gauge: Estimating Chain-of-Thought Length Ahead of Time in Large Multimodal Modelscited by 0
  17. 2026The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearningcited by 0
  18. 2026LLM Unlearning Evaluation with TRIAGEcited by 0
  19. 2026Steering by Influence: Curvature Aware Data Weighting for Activation Steeringcited by 0
  20. 2026REDaCT: Collapse-Free Unlearning in Large Reasoning Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter
  2. 2026Representation-Guided Parameter-Efficient LLM Unlearning
  3. 2026Unlearning in LLMs: Methods, Evaluation, and Open Challenges
  4. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMs
  5. 2026Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMs
  6. 2025BLUR: A Benchmark for LLM Unlearning Robust to Forget-Retain Overlap
  7. 2025LLM Unlearning on Noisy Forget Sets: A Study of Incomplete, Rewritten, and Watermarked Data
  8. 2024RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models