Due: 4 March 2019 at the beginning of class

• Read section 5.3 and 5.4 in Bengio (on D2L).
• Answer the following questions. Follow the general homework directions.
  • Train a Restricted Boltzman Machine with 4 visible nodes and 3 hidden nodes. Initialize the weight matrix and bias vectors with values drawn from N(0,1). Create an input vector of your choice. You may do this by hand, with existing code, or your own code.
  • Compare and contrast a shallow neural network with a deep neural network on the the Wisconsin Breast Cancer, Diabetes, Soybean, and Vote databases from the UCI ML Repository. These are also available from the WEKA data folder (breast-cancer.arff, diabetes.arff, soybean.arff, and vote.arff). Study the difference between various shallow and deep architectures. Present the confusion matrix and accuracy for each experiment. Discuss your results. There is existing neural network training packages in Matlab, Weka, and R.