We Should Distinguish Unlearning From Untraining
cs.LG
Submitted: 2026-04-09
Updated: 2026-11-09
Code: https://github.com/frankmcsherry/blog
Terminology
Sources
- Data Unlearning in Diffusion Models
- Information Complexity of Stochastic Convex Optimization: Applications to Generalization and Memorization
- To Each (Textual Sequence) Its Own: Improving Memorized-Data Unlearning in Large Language Models
- Open Problems in Machine Unlearning for AI Safety
- Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilities
- Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy and Research
- Unlearning Personal Data from a Single Image
- Do Unlearning Methods Remove Information from Language Model Weights?
- SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation
- Towards Adversarial Evaluations for Inexact Machine Unlearning
- Corrective Machine Unlearning
- Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy
- Unlearning or Obfuscating? Jogging the Memory of Unlearned LLMs via Benign Relearning
- Characterizing Structural Regularities of Labeled Data in Overparameterized Models
- Unifying Re-Identification, Attribute Inference, and Data Reconstruction Risks in Differential Privacy
- LLM Unlearning with LLM Beliefs
- The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
- Towards Safer Large Language Models through Machine Unlearning
- An Adversarial Perspective on Machine Unlearning for AI Safety
- Eight Methods to Evaluate Robust Unlearning in LLMs
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis