Research map / Continual learning / Online and replay-based
Mixed-Privacy Forgetting in Deep Networks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use a mixed-privacy model with nonlinear core weights and linear user weights | ResNet-50 image classification models · Caltech-256, MIT-67, Stanford Dogs, CUB-200, FGVC Aircrafts, and CIFAR-10 | test accuracy after forgetting | Helped |
| Use ML-Forgetting instead of Fisher noise or no forgetting procedure | ResNet-50 · 10% of training data forgotten in 10 sequential requests | readout-function similarity to retraining without forgotten data | Helped |
| Use 2–3 data passes for forgetting instead of retraining for 50 epochs | ML-Forgetting · Caltech-256 and FGVC Aircrafts | speedup per forgetting request: 16-25× | Helped |
| Compute the residual gradient completely rather than stochastically (ablation) | ML-Forgetting · 10% forgotten through 10 requests | test error after forgetting | Helped |
| Compute the residual gradient completely rather than stochastically (ablation) | ML-Forgetting · 10% forgotten through 10 requests | remaining information | Helped |
| Add task-relevant user data to the core set (ablation) | ML-Forgetting · FGVC Aircrafts | test accuracy | Helped |
| Add task-relevant user data to the core set (ablation) | ML-Forgetting · FGVC Aircrafts and CUB-200 | remaining information | Helped |
| Increase the fraction of task-distribution data in the core set (ablation) | Mixed-Linear model · FGVC Aircrafts | test accuracy | Helped |
| Increase the forgetting noise (ablation) | ML-Forgetting · Caltech-256 and FGVC Aircrafts | remaining information | Helped |
| Increase the forgetting noise (ablation) | ML-Forgetting · Caltech-256 and FGVC Aircrafts | test error | Hurt |
| Run more optimization steps during forgetting (ablation) | ML-Forgetting · Caltech-256 and FGVC Aircrafts | remaining information | Helped |
| Increase L2 regularization (ablation) | ML-Forgetting · Image classification tasks | remaining information | Helped |
Closest papers by meaning
- Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks (CVPR 2020 · similarity 0.91)
- Synthetic Forgetting Without Access: A Few-Shot Zero-Glance Framework for Machine Unlearning (AAAI 2026 · similarity 0.89)
- Distribution-Level Feature Distancing for Machine Unlearning: Towards a Better Trade-off Between Model Utility and Forgetting (AAAI 2025 · similarity 0.86)
- Continual Forgetting for Pre-trained Vision Models (CVPR 2024 · similarity 0.86)
- On Local Overfitting and Forgetting in Deep Neural Networks (AAAI 2025 · similarity 0.85)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.85)
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.