MNIST
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
MNIST Heldout Digit 5 1 (test)
0.939AUPRC
7
MNIST Heldout Digit 4 1 (test)
0.911AUPRC
7
MNIST Heldout Digit 1 (test)
68.4AUPRC
7
MNIST 5000 train, 8000 (test)
96.74Accuracy
7
MNIST Bernoulli shift
1.03Average Error
7
MNIST Linear shift
1.06Avg Error (%)
7
MNIST Square shift (test)
1.01Error Rate
7
MNIST Sine shift (test)
0.013Average Error (%)
7
MNIST 10k raw-pixel
68.7Accuracy
7
MNIST 50% symmetric noise (test)
94.38Accuracy (Test)
7
MNIST (train)
229.27Training Time (s)
7
MNIST 70% symmetric label noise (test)
91.61Accuracy
7
MNIST Wide first 1000 images (val)
-5.4Relative Verification Bound
7
MNIST Wide (val)
8.3Time (s)
7
MNIST Deep (val)
15.4Time (s)
7
MNIST FFNet (val)
5.2Execution Time (s)
7
Permuted MNIST 9 (test)
83.56Accuracy
7
MNIST LeNet-300-100 (test)
1.41Test Error
7
mnist LIBSVM (test)
0.932Average AUC
7
MNIST 17
0.3Abs Estimation Error
7
Binarized MNIST
2.4Absolute Estimation Error (%)
7
MNIST3000
95.54Accuracy
7
MNIST 2000
95.35Accuracy
7
MNIST 1000
94.66Accuracy
7
MNIST
78.1NLL
7