Image Classification on MedMNIST PathMNIST (test)
86.63AccuracyFedORA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FedORAm (number of unlearning classes)=2, n (percentage of samples to unlearn)=50%2025.12 | 86.63 | — | — | — | |
| FedORAm (number of unlearning classes)=5, n (percentage of samples to unlearn)=50%2025.12 | 85.89 | — | — | — | |
| CVFUm (number of unlearning classes)=2, n (percentage of samples to unlearn)=50%2025.12 | 85.47 | — | — | — | |
| Retrainm (number of unlearning classes)=2, n (percentage of samples to unlearn)=50%2025.12 | 85.42 | — | — | — | |
| Retrainm (number of unlearning classes)=5, n (percentage of samples to unlearn)=50%2025.12 | 84.82 | — | — | — | |
| ICOm (number of unlearning classes)=2, n (percentage of samples to unlearn)=50%2025.12 | 84.42 | — | — | — | |
| Wasserstein FIObjective Function=Wasserstein FI-based regularizer, Backbone=ResNet-18, Number of repetitions=10, Input Resolution=28x282025.02 | 84.03 | 95.68 | 0.6425 | 8.56 | |
| GAm (number of unlearning classes)=2, n (percentage of samples to unlearn)=50%2025.12 | 83.85 | — | — | — | |
| ERMObjective Function=Empirical Risk Minimization, Backbone=ResNet-18, Number of repetitions=10, Input Resolution=28x282025.02 | 83.22 | 96.84 | 0.7686 | 8.55 | |
| CVFUm (number of unlearning classes)=5, n (percentage of samples to unlearn)=50%2025.12 | 83.02 | — | — | — | |
| ICOm (number of unlearning classes)=5, n (percentage of samples to unlearn)=50%2025.12 | 82.41 | — | — | — | |
| GAm (number of unlearning classes)=5, n (percentage of samples to unlearn)=50%2025.12 | 79.48 | — | — | — |