Interpretable Network Architectures for Machine Learning