Unlearning Attack Success on MNIST, Fashion-MNIST, SVHN, and CIFAR-10 Average (test)
90.31NG Success RateModel Parameter Difference
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Model Parameter DifferenceAttack Method=Difference, Classification Method=Decision Tree, Unlearning Scenario=Single-Class2026.04 | 90.31 | 90.31 | 92.5 | 90.93 | 93.75 | 91.56 | |
| Model Parameter Dot ProductAttack Method=Dot Product, Classification Method=Youden’s Threshold, Unlearning Scenario=Single-Class2026.04 | 81.5 | 84.81 | 89.06 | 84.75 | 90.75 | 86.17 | |
| Model Parameter Dot ProductAttack Method=Dot Product, Classification Method=K-Means, Unlearning Scenario=Single-Class2026.04 | 80.12 | 83.5 | 88.31 | 83.93 | 90.25 | 85.22 |