Image Classification of handwritten digits on the famous MNIST dataset.
Database of handwritten digits - size normalized and centered in 28x28 fixed size greyscale images Using the MNIST dataset as provided by torchvision. 60,000 examples (and additionally 10,000 for testing)
A simple 2-layer NN, the size of the hidden dimension and the activation function are customizable.
LeCun et al. 1998 reported test error rate for various neural networks (an overview can be found on the MNIST website). Comparable networks have an error rate of roughly 1% to 5%