A New Type of Neural Network Is More Interpretable