Research map / Federated learning and privacy / Personalized and heterogeneous FL
Towards Privacy-Guaranteed Label Unlearning in Vertical Federated Learning: Few-Shot Forgetting Without Disclosure
Hanlin Gu, Hong Xi Tae, Lixin Fan, Chee Seng Chan
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 few-shot manifold mixup unlearning instead of Fisher forgetting | label unlearning in VFL · CIFAR10 (ResNet18) | accuracy on D^r (%): 54.4 -> 89.29 | Improved |
| use manifold mixup unlearning instead of Boundary Unlearning | label unlearning in VFL · ModelNet (ResNet18) | accuracy on D^r (%): 24.07 -> 87.69 | Improved |
| use manifold mixup unlearning instead of SSD | label unlearning in VFL · Brain Tumor MRI (ResNet18) | accuracy on D^r (%): 85.93 -> 93.79 | Improved |
| manifold mixup unlearning versus retraining | label unlearning in VFL · MNIST (ResNet18) | accuracy on D^r (%): 99.33 -> 99.05 | Hurt |
| add manifold mixup to few-shot gradient ascent (GA-S) (ablation) | few-shot gradient ascent · CIFAR10 (ResNet18) | unlearning of y^u and D^r accuracy | Mixed |
| add remained accuracy recovery module (ablation) | GA-S + manifold mixup · CIFAR10 (ResNet18) | D^r accuracy | Improved |
| few-shot gradient ascent without mixup (GA-S) instead of all-data GA-A (ablation) | gradient ascent · CIFAR10 (ResNet18) | unlearning of y^u | Hurt |
| vary mixup coefficient lambda (0.25 vs 0.5) (ablation) | manifold mixup unlearning · CIFAR10 (ResNet18) | unlearning effectiveness | No clear effect |
| increase number of passive parties (ablation) | manifold mixup unlearning · CIFAR10 (ResNet18) | D^r, y^u accuracy and ASR | No clear effect |
| use manifold mixup unlearning instead of retraining for process privacy | label unlearning in VFL · CIFAR10/100 (ResNet18) | membership leakage rate (%): 100 -> 14.38 | Improved |
| manifold mixup unlearning vs SSD on ASR | label unlearning in VFL · ModelNet and Brain MRI (ResNet18) | ASR | Mixed |
| manifold mixup unlearning vs Amnesiac on two-label ASR | label unlearning in VFL · CIFAR10 two-label (ResNet18) | ASR (%): 1.84 vs 24.21 | Hurt |
Closest papers by meaning
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.89)
- Text-Enhanced Data-free Approach for Federated Class-Incremental Learning (CVPR 2024 · similarity 0.86)
- Exploring Vacant Classes in Label-Skewed Federated Learning (AAAI 2025 · similarity 0.86)
- Exploiting Label Skews in Federated Learning with Model Concatenation (AAAI 2024 · similarity 0.86)
- Federated Unlearning with Gradient Descent and Conflict Mitigation (AAAI 2025 · similarity 0.86)
- Complementary Knowledge Distillation for Robust and Privacy-Preserving Model Serving in Vertical Federated Learning (AAAI 2024 · similarity 0.86)
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.
