CIFAR
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
CIFAR-100 90% class deletion
60.2UA
8
CIFAR-100 class-fraction-to-forget = 0.5
79Unlearning Accuracy (UA)
8
CIFAR-100 class 73 (shark) 50% deletion
74.6Utility Accuracy (UA)
8
CIFAR-100 apple class 50% deletion
94.4MIA Score
8
CIFAR-100 apple class 50% deletion (retain)
99.8RA
8
CIFAR-100 50% deletion of class 36 hamster (test)
95.9Utility Accuracy (UA)
8
CIFAR10 standard (test)
95.73Accuracy
8
CIFAR-10 L-infinity (test)
87.14Clean Accuracy
8
CIFAR-10
62.16OOD AUROC (sigma=0.05)
8
CIFAR-100 CIFAR-10 (test)
43.98FPR@95
8
CIFAR-100 Texture (test)
22.93FPR@95%
8
CIFAR-10 vs OOD Average (Mean) (test)
7.9FPR@95% TPR
8
CIFAR-100 ID vs SVHN TinyImageNet OOD
77.9AUPR (SVHN)
8
CIFAR-10-LT vs CIFAR-100
76.52AUPR
8
CIFAR-10 LT
91.9Misclassification AUPR
8
CIFAR-100 Non-IID
56.09Accuracy (Group 95-100%)
8
CIFAR-100 IID
60.67Accuracy (Group [95%, 100%])
8
CIFAR-10
95.34ID Accuracy
8
CIFAR10
91.57Task Accuracy
8
CIFAR-100
62.7Accuracy Retention
8
CIFAR-100
67.4Accuracy
8
CIFAR-10 (test)
51.09Accuracy (eps=0.5)
8
CIFAR10 Weak Non-IID
85.38Accuracy
8
CIFAR10 Strong Non-IID
82.57Accuracy
8
CIFAR10
-13.5SSIM
8