/CIFAR10_WideResNet22

Wide ResNet 22 model trained on the CIFAR10 image dataset.

Primary LanguageJupyter Notebook

CIFAR10_WideResNet22

Wide ResNet 22 model trained on the CIFAR10 image dataset.

  • Data normalization, data augmentation (4px padding, random 32x32 cropping, random horizontal flipping (p=0.5)).
  • Gradient clipping, Batch normalization, weight decay, learning rate scheduling, residual connections. Model trained for 20 epochs on a GPU achieves 93% test accuracy.