Use a convolutional neural network to classify handwritten digits from the MNIST dataset. A variety of architectures and hyperparameters were tested, and the best model (submitted to Kaggle) used an ensemble of ResNet-type models. The accuracy on the test set for this model was 99.585%.