CIFAR
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
CIFAR10 (test)
39Communication Rounds
50
CIFAR100
88.92Accuracy
50
CIFAR-100 (10000 labels)
8.99Error Rate
50
CIFAR+50
97.3AUROC
50
CIFAR-10 (test)
2.43NLL (bits/dim)
50
CIFAR 5+1
74.9Total Average Online Task Accuracy
49
CIFAR Random Label
98.42Total Average Online Task Accuracy
49
CIFAR-100
100True Positive Rate (TPR)
49
CIFAR-10
100AUC
49
CIFAR-100
96.3AUC
49
CIFAR100-F fine-grained
76Accuracy
49
CIFAR10 (train)
89.02Precision
48
CIFAR-10
10.9FID
48
CIFAR-10
49.71PGD Accuracy (L-inf)
48
CIFAR-100 1 class
76.01Retain Accuracy
48
CIFAR-10 1 class
94.74Retain Accuracy
48
CIFAR-10
0.969AUC
48
CIFAR-100
82.86ToW (High)
48
CIFAR-C-100 (9 held-out corruption types)
84.88Accuracy
48
CIFAR-C-100 10 training corruption types (test)
91.26Accuracy
48
CIFAR-10 bird, Class 2 (test)
100Forgetting Accuracy (Class)
48
CIFAR-10 32x32 with ReFlow (test)
3.74FID
48
CIFAR-10 (test)
84.72Natural Accuracy
48
CIFAR100
0.23Training Loss
48
CIFAR-100
787,542Noise Scale
48