mlsanityData to 7 Oct 2026

Erasing Conceptual Knowledge from Language Models

arXiv 2410.02760 ↗First posted 3 Oct 2024cs.CL cs.LGcited by 34

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

In this work, we introduce Erasure of Language Memory (ELM), a principled approach to concept-level unlearning that operates by matching distributions defined by the model's own introspective classification capabilities. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the language model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative evaluation reveals that ELM-modified models achieve near-random performance on assessments targeting erased concepts, while simultaneously preserving generation coherence, maintaining benchmark performance on unrelated tasks, and exhibiting strong robustness to adversarial attacks. Our code, data, and trained models are available at https://elm.baulab.info

Key facts. Cited by 34 papers in the map (data through 7 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 36 of its 48 references that are in the map. First posted on arXiv 3 Oct 2024.

Builds on

36 of 48 references.

Closest ancestors · 10

Same direction, newest first

  1. 2024Alternate Preference Optimization for Unlearning Factual Knowledge in Large Language Models20 Sep
  2. 2024Eight Methods to Evaluate Robust Unlearning in LLMs26 Feb
  3. 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
  4. 2023Towards Machine Unlearning Benchmarks: Forgetting the Personal Identities in Facial Recognition Systems3 Nov
  5. 2023Large Language Model Unlearning14 Oct
  6. 2023Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampening15 Aug
  7. 2023Towards Unbounded Machine Unlearning20 Feb
  8. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Models4 Oct
  9. 2022Quark: Controllable Text Generation with Reinforced Unlearning26 May
  10. 2022Towards Adversarial Evaluations for Inexact Machine Unlearning17 Jan

Other references · 26

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2022Classifier-Free Diffusion Guidancecited by 5,717
  5. 2023Mistral 7Bcited by 3,065
  6. 2016Pointer Sentinel Mixture Modelscited by 2,992
  7. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  8. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  9. 2024Qwen2 Technical Reportcited by 2,076
  10. 2021A General Language Assistant as a Laboratory for Alignmentcited by 1,023
  11. 2023Dissecting Recall of Factual Associations in Auto-Regressive Language Modelscited by 512
  12. 2023Zephyr: Direct Distillation of LM Alignmentcited by 498
14 more
  1. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 488
  2. 2023Unified Concept Editing in Diffusion Modelscited by 377
  3. 2023Ablating Concepts in Text-to-Image Diffusion Modelscited by 312
  4. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Modelscited by 297
  5. 2023Selective Amnesia: A Continual Learning Approach to Forgetting in Deep Generative Modelscited by 175
  6. 2020Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observationscited by 171
  7. 2024Tamper-Resistant Safeguards for Open-Weight LLMscited by 132
  8. 2023Stay on topic with Classifier-Free Guidancecited by 90
  9. 2024Fast Adversarial Attacks on Language Models In One GPU Minutecited by 85
  10. 2024Defending Against Unforeseen Failure Modes with Latent Adversarial Trainingcited by 82
  11. 2024Decoding-time Realignment of Language Modelscited by 67
  12. 2023An Emulator for Fine-Tuning Large Language Models using Small Language Modelscited by 61
  13. 2020Learn to Forget: Machine Unlearning via Neuron Maskingcited by 56
  14. 2024In-Context Editing: Learning Knowledge from Self-Induced Distributionscited by 33

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 1; 2025: 17; 2026: 16.

34 papers.

Sort:

By area:

  1. 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metricscited by 80
  2. 2025Measuring Chain of Thought Faithfulness by Unlearning Reasoning Stepscited by 43
  3. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMscited by 41
  4. 2025Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilitiescited by 36
  5. 2025ReLearn: Unlearning via Learning for Large Language Modelscited by 20
  6. 2025FADE: Why Bad Descriptions Happen to Good Featurescited by 13
  7. 2025Precise In-Parameter Concept Erasure in Large Language Modelscited by 11
  8. 2026Shaping capabilities with token-level data filteringcited by 11
  9. 2025AegisLLM: Scaling Agentic Systems for Self-Reflective Defense in LLM Securitycited by 9
  10. 2025OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Modelscited by 9
  11. 2025CRISP: Persistent Concept Unlearning via Sparse Autoencoderscited by 9
  12. 2025SteerVLM: Robust Model Control through Lightweight Activation Steering for Vision Language Modelscited by 8
22 more
  1. 2026The Story is Not the Science: Execution-Grounded Evaluation of Mechanistic Interpretability Researchcited by 6
  2. 2025Prompt Attacks Reveal Superficial Knowledge Removal in Unlearning Methodscited by 5
  3. 2025LLM Unlearning Without an Expert Curated Datasetcited by 5
  4. 2026Gauss-Newton Unlearning for the LLM Eracited by 4
  5. 2025ZJUKLAB at SemEval-2025 Task 4: Unlearning via Model Mergingcited by 2
  6. 2026From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearningcited by 2
  7. 2026TRACER: Token ReAssignment for Concept ERasure in Generative Recommendationcited by 2
  8. 2025LLM Unlearning Under the Microscope: A Full-Stack View on Methods and Metricscited by 1
  9. 2026Steerability of Instrumental-Convergence Tendencies in LLMscited by 1
  10. 2026Obliviate: Erasing Concepts from Autoregressive Image Generation Modelscited by 1
  11. 2024Targeted Angular Reversal of Weights (TARS) for Knowledge Removal in Large Language Modelscited by 0
  12. 2025REMIND: Input Loss Landscapes Reveal Residual Memorization in Post-Unlearning LLMscited by 0
  13. 2025RippleBench: Capturing Ripple Effects Using Existing Knowledge Repositoriescited by 0
  14. 2026Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Modelscited by 0
  15. 2026Beyond Dominant Patches: Spatial Credit Redistribution For Grounded Vision-Language Modelscited by 0
  16. 2026TEMPO: Temporal Enforcement via Mode-Separated Policy Optimization for Trustworthy LLM Backtestingcited by 0
  17. 2026DualOptim+: Bridging Shared and Decoupled Optimizer States for Better Machine Unlearning in Large Language Modelscited by 0
  18. 2026Don't Forget Your Embeddings: Robust Knowledge Erasure via Precise Editing of Embeddingscited by 0
  19. 2026MPSelectTune: Prompt-type Selection for Fine-tuning improves Concept Unlearning in LLMscited by 0
  20. 2026What to Forget in Unlearning? Forget Set Curation for Language Modelscited by 0
  21. 2026Spectral Saliency for Machine Unlearningcited by 0
  22. 2026Faithful Dual-constrained Erasure for Robust LLM Safety Alignmentcited by 0