Exercise in Convolutional Neural Networks for Classification * Choose a dataset, e g MNIST (0-1), MNIST (0-9), OCR (a-z) * Choose a framework * Choose a network topology * Train a convolutional neural network for classification * Think about a few or all of the following questions. Evaluate how well it performs. Try to run it on a few images, where you examine the values of a few different layers. Study the firs convolutional kernels. What do they look like? Study the last layer before classification. How do the values there correlate with the different classes? * One choice could be to run (Matconvnet) and do OCR. Read the following homepage http://www.robots.ox.ac.uk/~vgg/practicals/cnn/ And do exercise 4 on that homepage. * Another choice could be to run tensorflow and do MNIST Use the following tutorial https://www.codacy.com/app/hunkim/TensorFlow-Tutorials/files?bid=3237149 and run 05_convolutional_net.py I haven’t tried it myself yet, but I think it might be a good choice.