Model Editing and Unlearning

Model editing is where a model, such as a large language model, is “edited” to change the facts in the model. Machine unlearning is where a trained model is adjusted to “remove” one or more datapoints that were used to train the model, so that it behaves like a model that was trained without those datapoints. The datapoints to remove can either be specific datapoints from the training set, or classes of datapoints, such as all datapoints about bioweapons.

See also Knowledge Editing.

• Tan et al 2023 - Massive Editing for Large Language Models via Meta Learning
• Zhang et al 2024 - Towards General Conceptual Model Editing via Adversarial Representation Engineering

• Nguyen et al 2022 - A Survey of Machine Unlearning
• Xu et al 2023 - Machine Unlearning: A Survey
• Wang et al 2024 - Machine Unlearning: A Comprehensive Survey
• Liu et al 2024 - Machine Unlearning in Generative AI: A Survey
• Paper lists
  • Awesome LLM Unlearning
• For NLP or LLMs
  • Liu et al 2024 - Rethinking Machine Unlearning for Large Language Models (This is also a survey paper.)
  • Geng et al 2025 - A Comprehensive Survey of Machine Unlearning Techniques for Large Language Models

• Bourtoule et al 2019 - Machine Unlearning

• Eldan & Russinovich 2023 - Who’s Harry Potter? Approximate Unlearning in LLMs
• Yao et al 2023 - Large Language Model Unlearning
• Maini et al 2024 - TOFU: A Task of Fictitious Unlearning for LLMs
• Liu et al 2024 - Rethinking Machine Unlearning for Large Language Models
• Thaker et al 2024 - Guardrail Baselines for Unlearning in LLMs
• Gandikota et al 2024 - Erasing Conceptual Knowledge from Language Models
• Gur-Arieh et al 2025 - Precise In-Parameter Concept Erasure in Large Language Models

• Guo et al 2024 - Certified Data Removal from Machine Learning Models

• Privacy
• Knowledge Editing