Federated Data-level Unlearning on MNIST LeNet5 (test)
0.9918Pretrain Accuracyf-FUM (KL-KL)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| f-FUM (KL-KL)N (Number of clients)=10, Removed (%)=2, divergence=KL-KL2026.02 | 0.9918 | 0.774 | 0.5989 | |
| f-FUM (KL-JS)N (Number of clients)=10, Removed (%)=2, divergence=KL-JS2026.02 | 0.9918 | 0.7786 | 0.5933 | |
| f-FUM (KL-chi^2)N (Number of clients)=10, Removed (%)=2, divergence=KL-chi^22026.02 | 0.9918 | 0.7756 | 0.5978 | |
| NoTN (Number of clients)=10, Removed (%)=22026.02 | 0.9918 | 0.7292 | 0.5011 | |
| HalimiN (Number of clients)=10, Removed (%)=22026.02 | 0.9918 | 0.7772 | 0.5967 | |
| f-FUM (KL-KL)N (Number of clients)=5, Removed (%)=2, divergence=KL-KL2026.02 | 0.9893 | 0.7766 | 0.5933 | |
| f-FUM (KL-JS)N (Number of clients)=5, Removed (%)=2, divergence=KL-JS2026.02 | 0.9893 | 0.7772 | 0.58 | |
| f-FUM (KL-chi^2)N (Number of clients)=5, Removed (%)=2, divergence=KL-chi^22026.02 | 0.9893 | 0.7752 | 0.61 | |
| NoTN (Number of clients)=5, Removed (%)=22026.02 | 0.9893 | 0.7296 | 0.5089 | |
| HalimiN (Number of clients)=5, Removed (%)=22026.02 | 0.9893 | 0.7788 | 0.6189 |