Explaining Neural Networks

Dan discusses using mutual information to show the relationship between the diagnosis and explanation branches in a neural network. Kyle explores the challenge of making the explanation satisfying to domain experts without requiring too many labels.