Digit Classification research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

Rotating MNIST 45°

PCIDA

90.1Accuracy

Feb 26, 2026

Rotating MNIST 0°

CIDA

0.991Accuracy

Feb 26, 2026

MNIST -> USPS (partial protocol)

Generate To Adapt

92.8Accuracy

Feb 26, 2026

MultiMNIST (test)

MGDA-UB

97.26Left Digit Accuracy

Feb 26, 2026

Shuffled MNIST (val)

Mean Entropy

1Accuracy (Kendall's tau)

Feb 26, 2026

Colored MNIST (val)

Mean Entropy

1Validation Accuracy (Kendall's tau)

Feb 26, 2026

MNIST standard (test)

VMI + LOMMA

100Attack Accuracy

Feb 27, 2026

AV-MNIST Vision modality standard (test)

(xo, xm)

71.32Accuracy

Feb 26, 2026

MNIST-M to MNIST

DRANet

99.3Accuracy

Feb 26, 2026

MNIST -> USPS full dataset

PixelDA

95.9Accuracy

Feb 26, 2026

MNIST Manifold (test)

PTCNet

97.96Success Rate

Jun 5, 2026

MNIST Euclidean (test)

Traditional

98.85Success Rate

Jun 5, 2026

MNIST

—

—KNN Distance

Feb 26, 2026