Machine Learning Practical - Coursework 2 Report: Analysing problems with the VGG deep neural network architectures (with 8 and 38 hidden layers) on the CIFAR100 dataset by monitoring gradient flow during training. And exploring solutions using batch normalization and residual connections.

Dropout vs. batch normalization: effect on accuracy, training and inference times - code for the paper