Federated Unlearning on MNIST (val)
99.14Pretrain Accf-FUM (KL-KL)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| f-FUM (KL-KL)N=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 99.16 | |
| f-FUM (KL-JS)N=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 98.99 | |
| f-FUM (KL-χ2)N=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 99.21 | |
| NoTN=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 82.6 | |
| HalimiN=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 99.21 | |
| MoDEN=10, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 99.14 | 98.74 | |
| f-FUM (KL-KL)N=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 98.88 | |
| f-FUM (KL-JS)N=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 98.81 | |
| f-FUM (KL-χ2)N=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 98.95 | |
| NoTN=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 89.21 | |
| HalimiN=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 98.81 | |
| MoDEN=5, Backbone=LeNet5, Unlearning Scenario=Client-level (80% backdoored data removed)2026.02 | 98.98 | 98.49 |