CIFAR
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
CIFAR10
98.4Clean Accuracy
20
CIFAR-10 epsilon=2/255 (first 200 examples)
54Verified Accuracy
20
CIFAR-100 5 tasks (test)
69.9AN
20
CIFAR-100 varying-accuracy synthetic expert (test)
18.09Error Rate
20
CIFAR-100 Overlapped Animal Expert
15.1Error Rate
20
CIFAR-10 cat-dog (test)
0.68Posterior Expected Log Loss
20
CIFAR-10 deer-horse (test)
0.616Log Loss
20
CIFAR-10 airplane-automobile (test)
0.466Posterior Expected Log Loss
20
CIFAR-10
68.7Pretrain Accuracy
20
CIFAR-100 (test)
53.2Efficacy AUC
20
CIFAR-10 n=10K
88.96Accuracy
20
Cifar-10 (test)
99Accuracy
20
CIFAR-10 (train)
90.07MSE
20
CIFAR-10
0.885AUC
20
CIFAR-100 60 base classes 5-way 5-shot Session 8
86.06Accuracy
20
CIFAR-100 60 base classes 5-way 5-shot Session 7
87.15Accuracy
20
CIFAR-100 60 base classes, 5-way 5-shot, Session 6
87.53Accuracy
20
CIFAR-100 60 base classes 5-way 5-shot Session 4
87.83Accuracy
20
CIFAR-100 60 base classes 5-way 5-shot Session 2
88.97Accuracy
20
CIFAR-100 60 base classes 5-way 5-shot Session 1
89.39Accuracy
20
CIFAR-100 60 base classes, 5-way 5-shot, Session 0
90.88Accuracy
20
CIFAR-10 10/1 (test)
74.2Accuracy
20
CIFAR-10 class conditional 32x32
1.77FID
20
CIFAR-10 1.0 (test)
94.04BadNet C-Acc
20
CIFAR-10 Instance-Dependent Noise 40% (test)
90.6Accuracy
20