Machine Learning Breakthrough? – Sample instead of using algorithmic reduce. Keep matrices square from the start to the finish then sample out [:,:10]. Wow 98.5 on MNIST using an MLP. So the size of all the layers are 784*784 as the input size.

Machine Learning Breakthrough? – Sample instead of using algorithmic reduce. Keep matrices square from the start to the finish then sample out [:,:10]. Wow 98.5 on MNIST using an MLP. So the size of all the layers are 784*784 as the input size.