Research map / Learning theory and algorithms / Prediction and decision losses
Challenging Forgets: Unveiling the Worst-Case Forget Sets in Machine Unlearning
Chongyu Fan, Jiancheng Liu, Alfred Hero, Sijia Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| evaluate unlearning on the bilevel-optimized worst-case forget set instead of a random forget set | Retrain (exact unlearning) · CIFAR-10, ResNet-18 | UA: 5.28 -> 0.00 (10% forgetting) | Improved |
| forget worst-case set instead of random set | Retrain · CIFAR-10, ResNet-18 | TA: 94.38 -> 94.66 (10%); 94.04 -> 94.60 (20%) | Improved |
| evaluate relabeling-based methods (RL, BE, BS, SalUn) on worst-case forget set | RL, BE, BS, SalUn · CIFAR-10, 10% forgetting | Avg. gap vs Retrain: SalUn 1.89 -> 24.51; BE 4.28 -> 34.70 | Hurt |
| evaluate relabeling-free methods on worst-case forget set | FT, EUk, CFk, Scrub, l1-sparse · CIFAR-10, 10% forgetting | Avg. gap vs Retrain: Scrub 9.91 -> 0.82; FT 2.00 -> 1.37 | Improved |
| worst-case forget set reduces evaluation variance versus random forget set | Scrub · CIFAR-10, 10% forgetting | UA std: 12.42 ±19.82 -> 0.01 ±0.01 | Improved |
| train on complement of worst-case forget set as a coreset | ResNet-18 · CIFAR-10 / CIFAR-100 | test accuracy | Improved |
| forget worst-case prompt set instead of random prompts with ESD | Stable Diffusion on UnlearnCanvas · UnlearnCanvas prompt-wise forgetting | qualitative unlearning success | Hurt |
Closest papers by meaning
- What makes unlearning hard and what to do about it (NeurIPS 2024 · similarity 0.93)
- Not All Wrong is Bad: Using Adversarial Examples for Unlearning (ICML 2025 · similarity 0.93)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.92)
- Machine Unlearning for Image-to-Image Generative Models (ICLR 2024 · similarity 0.91)
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 2025 · similarity 0.91)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
