Images as numbers

For tasks such as these, we need deep networks with multiple hidden layers if we are hoping to learn any representative features for our output classes. We also need a nice dataset to practice our understanding and familiarize ourselves with the tools we will be using to design our intelligent systems. Hence, we come to our first hands-on neural network task as we introduce ourselves to the concepts of computer vision, image processing, and hierarchical representation learning. Our task at hand is to teach computers to read numbers not as 0 and 1s, as they already do, but more in the manner of how we would read digits that are composed by our own kin. We are speaking of handwritten digits, and for this task, we will be using the iconic MNIST dataset, the true hello world of deep learning datasets. For our first example, there are good theoretical and practical...