Research map / Label-efficient and robust learning / Recognition and long-tail
Designing to Forget: Deep Semi-parametric Models for Unlearning
Amber Yijia Zheng, Yu-Shan Tai, Raymond A. Yeh
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 |
|---|---|---|---|
| add label-permutation augmentation (ablation) | SPM generation · CIFAR-10 class-wise forgetting (Dog) | Delta UA: 83.61 -> 0.04 | Improved |
| add label-permutation augmentation (ablation) | SPM generation · CIFAR-10 class-wise forgetting (Horse) | Delta UA: 60.89 -> 0.08 | Improved |
| add label-permutation augmentation (ablation) | SPM generation · CIFAR-10 class-wise forgetting (Cat) | Delta FID_R: 2.68 -> 0.50 | Improved |
| add label-permutation augmentation (ablation) | SPM generation · CIFAR-10 class-wise forgetting (Automobile) | FID_O: 2.88 -> 1.97 | Improved |
| use clustering (SPM-C) instead of retrieval (SPM-R) for the set | SPM · ImageNet-1K classification | top-1 accuracy: 59.9 -> 67.1 | Improved |
| use a larger inputted set at test time (10% -> 100%) | SPM-C · ImageNet-1K classification | top-1 accuracy: 66.6 -> 67.1 | Improved |
| use a larger inputted set |T| (10 -> 1024) | SPM generation · CIFAR-10 class-conditioned generation | FID: 7.71 -> 7.04 | Improved |
| use a larger inputted set |T| (10 -> 1024) | SPM generation · CIFAR-10 class-conditioned generation | runtime (s): 52 -> 1486 | Hurt |
| replace parametric ResNet18 with SPM-C | classification · ImageNet-1K classification | top-1 accuracy: 68.93 -> 67.1 | Hurt |
| use class-aware GNN as non-parametric module | SPM · CIFAR-10 classification | accuracy: 94.1 -> 94.4 | Improved |
| replace gradient-based unlearning (SalUn) with test-time deletion on SPM | unlearning · ImageNet-1K 1-class unlearning | PG_S: 0.42 -> 0.12 | Improved |
| replace gradient-based unlearning (SA) with test-time deletion on SPM | unlearning · CIFAR-10 generation (Cat) | FID_O: 32.54 -> 1.85 | Improved |
Closest papers by meaning
- Machine Unlearning for Image-to-Image Generative Models (ICLR 2024 · similarity 0.90)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · similarity 0.89)
- Not All Wrong is Bad: Using Adversarial Examples for Unlearning (ICML 2025 · similarity 0.89)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.89)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.89)
- The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation (NeurIPS 2026 · similarity 0.89)
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.
