Classification on CIFAR100 (Baseline and Unlearned Metrics)
89.5Baseline mTAROKA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ROKAModel=ViT-base/16, Unlearning Method=Targeted, Task=Classification2026.02 | 89.5 | 89.87 | 15.89 | 88.44 | |
| ROKAModel=ViT-base/16, Unlearning Method=Non-Targeted, Task=Classification2026.02 | 88.17 | 90.03 | 24.01 | 89.79 | |
| ROKAModel=DeiT-tiny/16, Unlearning Method=Targeted, Task=Classification2026.02 | 73 | 74.5 | 0.1 | 72.42 | |
| ROKAModel=DeiT-tiny/16, Unlearning Method=Non-Targeted, Task=Classification2026.02 | 69.2 | 75.32 | 23.2 | 65.03 | |
| ROKAModel=CLIP-large/14, Unlearning Method=Non-Targeted, Task=Zero-shot Classification2026.02 | 63.58 | — | 10 | 63.09 | |
| ROKAModel=CLIP-large/14, Unlearning Method=Targeted, Task=Zero-shot Classification2026.02 | 62.3 | 65.21 | 0 | 84.302 | |
| ROKAModel=CLIP-base/32, Unlearning Method=Targeted, Task=Zero-shot Classification2026.02 | 62 | 54.72 | 0 | 80.57 | |
| ROKAModel=CLIP-base/32, Unlearning Method=Non-Targeted, Task=Zero-shot Classification2026.02 | 55.6 | — | 8.6 | 52.84 |