Test increasingly complex PyTorch models (logistic regression, MLP, CNN) for image classification on the CIFAR10 dataset, with custom code for training routines, optimiser and evaluation metric.
Two different CNN architectures are trained:
- Model 1: Two convolutions with stride=1 + Max-Pooling + Dense net
- Model 2: Five convolutions with stride=2