WideResNet9 model on the CIFAR10 image dataset. Model trained for 30 epochs on a GPU achieves 93% test accuracy.

• Data Processing steps: data normalization (mean=0, sd=1), data augmentation (4px padding, random 32x32 cropping, p=0.5 random horizontal flip).
• Network architecture: using residual connections, batch normalization, ReLU activation.
• To accelerate learning: learning rate scheduling (one-cycle policy), gradient clipping.
• To prevent overfitting: weight decay (~l2 regularization).
• Optimizer: Adam.