Research map / Continual learning / Class-incremental
Attaining Class-Level Forgetting in Pretrained Model Using Few Samples
Pravendra Singh, Pratik Mazumder, Mohammed Asad Karim
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 excluded-class gradient-ascent loss to classification-only tuning (ablation) | ERwP · CIFAR-100 ResNet-56 | FA_e: 66.50% -> 0.00% | Improved |
| add knowledge distillation loss on remaining-class logits (ablation) | ERwP · CIFAR-100 ResNet-56 | CA_ne: 6.45% -> 69.32% | Improved |
| add knowledge distillation loss on remaining-class logits (ablation) | ERwP · CIFAR-100 ResNet-56 | FPA_e: 24.40% -> 47.84% | Hurt |
| use all identified relevant parameters instead of 25% of them (ablation) | ERwP · CIFAR-100 ResNet-56 | FPA_e: 60.35% -> 47.84% | Improved |
| use all identified relevant parameters instead of 50% of them (ablation) | ERwP · CIFAR-100 ResNet-56 | FPA_e: 53.7% -> 47.84% | Improved |
| set beta to 10 versus 8 (ablation) | ERwP · CIFAR-100 ResNet-20 | CA_ne: 66.03% -> 66.84% | Improved |
| set kappa to 2.0 versus 1.0 (ablation) | ERwP · CIFAR-100 ResNet-20 | CA_ne: 66.05% -> 66.84% | Improved |
| vary number of excluded classes from 5 to 40 (ablation) | ERwP · CIFAR-100 ResNet-56 | FA_e: 0.00% | No clear effect |
| use different augmentations (grayscale, flip, rotation, affine) to find relevant parameters (ablation) | ERwP · CIFAR-100 ResNet-56 | FA_e after pruning | No clear effect |
| apply ERwP with full training data instead of limited (ablation) | ERwP · CIFAR-100 ResNet-20 | FPA_e gap to FDR: 3.86% -> 2.79% | Improved |
| apply ERwP with full training data instead of limited (ablation) | ERwP · CIFAR-100 ResNet-56 | FPA_e gap to FDR: 2.44% -> 1.65% | Improved |
| fine-tune on remaining-class limited data only (baseline 8) vs ERwP | FOLNRC · CIFAR-100 ResNet-56 | FA_e: 66.90% -> 0.00% | Hurt |
Closest papers by meaning
- Partially Blinded Unlearning: Class Unlearning for Deep Networks from Bayesian Perspective (AAAI 2025 · similarity 0.86)
- Learning to Unlearn: Instance-Wise Unlearning for Pre-trained Classifiers (AAAI 2024 · similarity 0.85)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.85)
- Distribution-Level Feature Distancing for Machine Unlearning: Towards a Better Trade-off Between Model Utility and Forgetting (AAAI 2025 · similarity 0.85)
- Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks (CVPR 2025 · similarity 0.85)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.84)
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.
