Skip to main content

Training wide networks for MNIST with PyTorch

These days, deep networks are very popular. For MNIST dataset, pytorch has very simple script for training. But you can also train wide networks and achieve accuracy of 99.42 % (same as in DeeperThought and WideOpenThoughts configG). Here is a script for wide pytorch training. Testing accuracy progress (max is 99.42 %):
Or if you want to go even further, you can achieve similar with really wide networks.