Improving Accuracy of Small Interpretable Models by Learning the Training Distribution

Learning the training data distribution can significantly improve the accuracy of small interpretable models, making them competitive with specialized techniques.