Get Started
Home
Topics
Search
Library
Research questionHow can class unlearning be audited for recoverable forget-class structure with only white-box model access?A model can show low accuracy on a designated forget class while retaining representation structure that allows the class to be recovered. Auditing this distinction is difficult when source data and comparison checkpoints are unavailable.
AI
Computer Vision
Evaluation & Benchmarks
Machine Learning
Latest papersRecent research connected to this question, newest first.Source-Free Class Relearning: Diagnosing Forgetting in Class UnlearningThe source studies image-classification unlearning in a strictly source-free, white-box setting, using classifier-head updates and synthetic probes in representation space. Evidence covers CIFAR-10, CIFAR-100, and TinyImageNet with ResNet-18, ViT-B/16, and Swin-T across several unlearning methods; behavior beyond these settings is not established.research paper · Sep 2, 2026
Related questions
How can federated unlearning prevent classifier broadcasts from revealing and enabling reinsertion of deleted client data?Can black-box attackers identify and reconstruct prompts supposedly removed from language models without knowing them in advance?How can LLMs remove undesirable knowledge while preserving utility with limited retention and unlearning data?How can stateful LLM agents behave as if they never saw revoked information?