Research map / Federated learning and privacy / Differential privacy
BlindU: Blind Machine Unlearning Without Revealing Erasing Data
Weiqi Wang, Zhiyi Tian, Chenhan Zhang, Shui Yu
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 |
|---|---|---|---|
| use BlindU_w/o instead of VBU for backdoor removal | BlindU · CIFAR10 (EDR 6%) | accuracy on erased data: 7.92 -> 6.57 | Improved |
| use BlindU instead of VBU for model utility | BlindU_w · CIFAR10 (backdoor unlearning) | test accuracy: 65.45 -> 75.83 | Improved |
| use BlindU instead of VBU for model utility | BlindU_w · Adult (backdoor unlearning) | test accuracy: 64.64 -> 84.34 | Improved |
| use BlindU instead of HBFU on backdoor removal | BlindU_w/o · Adult | accuracy on erased data: 99.99 -> 7.25 | Improved |
| add DP noise to HBFU (DP-HBFU) (ablation) | HBFU · MNIST | model accuracy: 96.04 -> 75.94 | Hurt |
| add DP noise to BFU (DP-BFU) (ablation) | BFU · MNIST | model accuracy: 96.38 -> 81.73 | Hurt |
| add DP noise to VBU (DP-VBU) (ablation) | VBU · MNIST | model accuracy: 88.19 -> 48.02 | Hurt |
| increase DP sampling ratio from 20% to 60% (with replacement) (ablation) | BlindU_w · MNIST | backdoor accuracy: 3.31 -> 1.89 | Improved |
| increase DP sampling ratio from 20% to 60% (with replacement) (ablation) | BlindU_w · MNIST | model accuracy: 95.86 -> 94.56 | Hurt |
| sample without replacement instead of with replacement (ablation) | BlindU · MNIST (SR 60%) | backdoor accuracy: 1.89 -> 1.56 | Improved |
| sample without replacement instead of with replacement (ablation) | BlindU · MNIST (SR 60%) | epsilon: 0.597 -> 0.902 | Hurt |
| use ViT approximator instead of ResNet-18 | BlindU_w · MNIST | accuracy after unlearning: 94.55 -> 96.38 | Improved |
Closest papers by meaning
- The Utility and Complexity of In- and Out-of-Distribution Machine Unlearning (ICLR 2025 · similarity 0.92)
- A Certified Unlearning Approach without Access to Source Data (ICML 2025 · similarity 0.90)
- Leveraging Per-Instance Privacy for Machine Unlearning (ICML 2025 · similarity 0.90)
- Towards Source-Free Machine Unlearning (CVPR 2025 · similarity 0.90)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.90)
- Langevin Unlearning: A New Perspective of Noisy Gradient Descent for Machine Unlearning (NeurIPS 2024 · 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.
