Pith. sign in

REVIEW 19 cited by

Knowledge Unlearning for Mitigating Privacy Risks in Language Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2210.01504 v2 pith:TZFIMW5N submitted 2022-10-04 cs.CL

classification cs.CL
keywords privacydataunlearninglanguageknowledgemethodmodelsrisks
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Pretrained Language Models (LMs) memorize a vast amount of knowledge during initial pretraining, including information that may violate the privacy of personal lives and identities. Previous work addressing privacy issues for language models has mostly focused on data preprocessing and differential privacy methods, both requiring re-training the underlying LM. We propose knowledge unlearning as an alternative method to reduce privacy risks for LMs post hoc. We show that simply performing gradient ascent on target token sequences is effective at forgetting them with little to no degradation of general language modeling performances for larger LMs; it sometimes even substantially improves the underlying LM with just a few iterations. We also find that sequential unlearning is better than trying to unlearn all the data at once and that unlearning is highly dependent on which kind of data (domain) is forgotten. By showing comparisons with a previous data preprocessing method and a decoding method known to mitigate privacy risks for LMs, we show that unlearning can give a stronger empirical privacy guarantee in scenarios where the data vulnerable to extraction attacks are known a priori while being much more efficient and robust. We release the code and dataset needed to replicate our results at https://github.com/joeljang/knowledge-unlearning.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 19 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Space Complexity of Learning-Unlearning Algorithms

    cs.LG 2025-06 accept novelty 8.0 of 10

    The space complexity of machine unlearning for realizability testing is characterized by eluder dimension (central lower bound), star number (ticketed upper bound), and hollow star number (bounded deletions), separati...

  2. System-Aware Unlearning Algorithms: Use Lesser, Forget Faster

    cs.LG 2025-06 conditional novelty 7.0 of 10

    The paper introduces system-aware unlearning and gives the first exact unlearning algorithm for linear classification that stores a sublinear-size core set instead of the entire dataset.

  3. Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs

    cs.LG 2026-07 conditional novelty 6.0 of 10

    A special <|ino|> token trained into LLMs gates hazardous answers behind an assistant-side signal, beating unlearning and refusal baselines on safety-utility trade-offs.

  4. Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning

    cs.CL 2026-02 conditional novelty 6.0 of 10

    SFT models forget facts more stably than pretrained models, with 10-50% higher retention of unrelated knowledge when using the DUET benchmark of 28.6k Wikidata triplets.

  5. Distribution Preference Optimization: A Fine-grained Perspective for LLM Unlearning

    cs.LG 2025-10 conditional novelty 6.0 of 10

    DiPO is a distribution-level unlearning method that constructs preference distributions from the model's own high-confidence logits and achieves state-of-the-art forget quality on TOFU while preserving utility.

  6. Towards Evaluation for Real-World LLM Unlearning

    cs.AI 2025-08 conditional novelty 6.0 of 10

    DCUE evaluates LLM unlearning by comparing core-token confidence score distributions of the unlearned model and the original model, corrected by a validation set, using the Kolmogorov-Smirnov test.

  7. Enhancing Safe and Controllable Protein Generation via Knowledge Preference Optimization

    cs.AI 2025-07 conditional novelty 6.0 of 10

    A knowledge-graph-guided preference optimization framework that fine-tunes protein language models to generate fewer sequences similar to known harmful proteins.

  8. Towards a Principled Evaluation of Knowledge Editors

    cs.CL 2025-07 conditional novelty 6.0 of 10

    The choice of evaluation metric, generation length, and edit batch size changes the ranking of knowledge editors, and exact string matching produces false positives.

  9. Model Collapse Is Not a Bug but a Feature in Machine Unlearning for LLMs

    cs.LG 2025-07 conditional novelty 6.0 of 10

    A new method, Partial Model Collapse, iteratively fine-tunes an LLM on its own self-generated responses to conditionally collapse its output distribution on forget queries, removing private answers without the true la...

  10. Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputs

    cs.LG 2025-06 conditional novelty 6.0 of 10

    Unlearning in LLMs leaves detectable 'fingerprints' that let a simple classifier distinguish an unlearned model from its original, even on unrelated prompts.

  11. LLM Unlearning Should Be Form-Independent

    cs.CL 2025-06 conditional novelty 6.0 of 10

    Existing LLM unlearning is form-dependent; the new ORT benchmark measures this, and the training-free ROCR edit reduces it by redirecting concept representations.

  12. Understanding and Mitigating Cross-lingual Privacy Leakage via Language-specific and Universal Privacy Neurons

    cs.CL 2025-06 reject novelty 6.0 of 10

    Cross-lingual privacy leakage in LLMs is driven by a mix of language-universal and language-specific neurons, and deactivating those neurons lowers measured leakage by 23.3% to 31.6%.

  13. Module-Aware Parameter-Efficient Machine Unlearning on Transformers

    cs.LG 2025-08 conditional novelty 5.0 of 10

    MAPE-Unlearn uses Fisher-information-based scores and greedy search to select important heads and filters, then applies sparse unlearning updates, claiming improved efficacy-fidelity trade-offs on Transformers.

  14. Reliable Unlearning Harmful Information in LLMs with Metamorphosis Representation Projection

    cs.LG 2025-08 unverdicted novelty 5.0 of 10

    An LLM unlearning method that projects hidden states so harmful information is irreversibly removed while useful knowledge is preserved.

  15. Automating Evaluation of Diffusion Model Unlearning with (Vision-) Language Model World Knowledge

    cs.LG 2025-07 conditional novelty 5.0 of 10

    A new evaluation tool uses (vision-)language model world knowledge to rank nearby concepts and craft adversarial prompts, showing that diffusion unlearning is incomplete and that semantic similarity correlates with co...

  16. Step-by-Step Reasoning Attack: Revealing 'Erased' Knowledge in Large Language Models

    cs.CR 2025-06 reject novelty 4.0 of 10

    Step-by-step reasoning prompts can recover purportedly erased facts from unlearned LLMs, but the paper's quantitative evidence is internally inconsistent.

  17. Lifting Data-Tracing Machine Unlearning to Knowledge-Tracing for Foundation Models

    cs.CV 2025-06 conditional novelty 4.0 of 10

    A position paper urging a shift from data-tracing to knowledge-tracing machine unlearning for foundation models, supported by a CLIP case study that shows current methods struggle to generalize.

  18. UCD: Unlearning in LLMs via Contrastive Decoding

    cs.CL 2025-06 conditional novelty 4.0 of 10

    UCD steers an LLM away from forget-set content at inference time by mixing in the difference between forget-tuned and retain-tuned small models.

  19. iShumei-Chinchunmei at SemEval-2025 Task 4: A balanced forgetting and retention multi-task framework using effective unlearning loss

    cs.CL 2025-07 conditional novelty 3.0 of 10

    The authors propose Effective Unlearning Loss, the inverse of the standard next-token prediction loss, within a multi-task framework, and report a 5th-place finish at SemEval-2025 Task 4.

Pith tools