mlsanityData to 14 Oct 2026

From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning

arXiv 2601.04278 ↗First posted 7 Jan 2026cs.CL cs.AI cs.CRcited by 2

On the map: Synthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training

also in 1 more
Abstract (arXiv)

Although machine unlearning is essential for removing private, harmful, or copyrighted content from LLMs, current benchmarks often fail to faithfully represent the true ``forgetting scope'' learned by the model. We formalize two distinct unlearning granularities, domain-level and instance-level, and propose BiForget, an automated framework for synthesizing high-quality forget sets. Unlike prior work relying on external generators, BiForget exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting. Our experiments across diverse benchmarks show that it achieves a superior balance of relevance, diversity, and efficiency. Quantitatively, in the Harry Potter domain, it improves relevance by ~20 and diversity by ~0.05 while halving the total data size compared to SOTAs. Ultimately, it facilitates more robust forgetting and better utility preservation, providing a more rigorous foundation for evaluating LLM unlearning.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Self-rewarding alignment and self-training, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 24 of its 35 references that are in the map. First posted on arXiv 7 Jan 2026.

Builds on

24 of 35 references.

References · 24

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  3. 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,173
  4. 2019Machine Unlearningcited by 1,061
  5. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 533
  6. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  7. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 504
  8. 2023Jailbreaking ChatGPT via Prompt Engineering: An Empirical Studycited by 498
  9. 2019ClinicalBERT: Modeling Clinical Notes and Predicting Hospital Readmissioncited by 493
  10. 2018Texygen: A Benchmarking Platform for Text Generation Modelscited by 491
  11. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 465
  12. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Modelscited by 437
12 more
  1. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Modelscited by 261
  2. 2023In-Context Unlearning: Language Models as Few Shot Unlearnerscited by 224
  3. 2024Tamper-Resistant Safeguards for Open-Weight LLMscited by 138
  4. 2024An Adversarial Perspective on Machine Unlearning for AI Safetycited by 123
  5. 2024Large Language Model Unlearning via Embedding-Corrupted Promptscited by 116
  6. 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metricscited by 83
  7. 2024Position: LLM Unlearning Benchmarks are Weak Measures of Progresscited by 63
  8. 2024On Large Language Model Continual Unlearningcited by 52
  9. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMscited by 42
  10. 2024Erasing Conceptual Knowledge from Language Modelscited by 34
  11. 2024Quality Matters: Evaluating Synthetic Data for Tool-Using LLMscited by 19
  12. 2025LLM Unlearning Without an Expert Curated Datasetcited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Position: The Term "Machine Unlearning" Is Overused in LLMscited by 3
  2. 2026Harmonizing Multi-Objective LLM Unlearning via Unified Domain Representation and Bidirectional Logit Distillationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025LLM Unlearning on Noisy Forget Sets: A Study of Incomplete, Rewritten, and Watermarked Data
  2. 2026Forget Narrowly, Retain Broadly: Unlearning as an Asymmetric Generalization Problem
  3. 2024Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference
  4. 2025BLUR: A Benchmark for LLM Unlearning Robust to Forget-Retain Overlap
  5. 2026Unlearning in LLMs: Methods, Evaluation, and Open Challenges
  6. 2026Confess What You Know: Forget-Set Misalignment with Model Knowledge in LLM Unlearning
  7. 2025The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluation
  8. 2024Towards Robust Evaluation of Unlearning in LLMs via Data Transformations