Research map / Federated learning and privacy / Differential privacy
Fully Decentralized Certified Unlearning
Hithem Lamri, Michail Maniatakos
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 |
|---|---|---|---|
| route to the unlearning client with p=1.0 instead of p=0.1 (continuous unlearning) (ablation) | RR-DU · MNIST | test accuracy (%): 99.18 -> 10.10 | Hurt |
| route to the unlearning client with p=1.0 instead of p=0.1 (continuous unlearning) (ablation) | RR-DU · CIFAR-10 | test accuracy (%): 88.30 -> 17.74 | Hurt |
| route to the unlearning client with p=1.0 instead of p=0.1 (ablation) | RR-DU · MNIST | backdoor accuracy (%): 10.33 -> 0.00 | Improved |
| use p=0.1 targeted unlearning instead of p=0 finetuning (ablation) | RR-DU · CIFAR-10 | backdoor accuracy (%): 10.93 -> 11.20 | No clear effect |
| use p=0.1 targeted unlearning instead of p=0 finetuning (ablation) | RR-DU · MNIST | test accuracy (%): 99.18 -> 99.18 | No clear effect |
| use RR-DU instead of finetuning after removing poisoned data | RR-DU · MNIST | backdoor accuracy (%): 18 -> 10 | Improved |
| use RR-DU instead of decentralized DP (DDP) | RR-DU · MNIST | backdoor accuracy (%): 35 -> 10 | Improved |
| use RR-DU instead of DP-SGD | RR-DU · MNIST | backdoor accuracy (%): 60 -> 10 | Improved |
| use RR-DU instead of DDP | RR-DU · MNIST | clean test accuracy (%): 96.7 -> 99.1-99.2 | Improved |
| use RR-DU instead of DP-SGD and DDP | RR-DU · CIFAR-10 | clean test accuracy (%): 50-55 -> 88-89 | Improved |
| use RR-DU instead of finetuning | RR-DU · CIFAR-10 | backdoor accuracy (%): 25-30 -> 10 | Improved |
Closest papers by meaning
- The Utility and Complexity of In- and Out-of-Distribution Machine Unlearning (ICLR 2025 · similarity 0.90)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.90)
- Rewind-to-Delete: Certified Machine Unlearning for Nonconvex Functions (NeurIPS 2025 · similarity 0.89)
- Leveraging Per-Instance Privacy for Machine Unlearning (ICML 2025 · similarity 0.89)
- Langevin Unlearning: A New Perspective of Noisy Gradient Descent for Machine Unlearning (NeurIPS 2024 · similarity 0.89)
- PDUDT: Provable Decentralized Unlearning under Dynamic Topologies (ICML 2025 · similarity 0.88)
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.
