Skip to main content

Table 2 Classification accuracy (%) for GCCS with ResNet-18 as a function of the training batch size, evaluated on CIFAR10 and MNIST

From: Gaussian class-conditional simplex loss for accurate, adversarially robust deep classifier training

Training Batch Size

50

100

200

400

CIFAR10

77.33

82.11

82.97

82.93

MNIST

95.14

99.35

99.58

99.56