LIVE: A Local Interpretable model-agnostic Visualizations and Explanations

Deep learning has been successfully applied to various types of classification and prediction tasks including images, text, and tabular data, however interpretability of deep learning models still remains challenges, which becomes a limiting factor for further applications.Most of current interpretable models are limited to how to interpret the deep learning models. In this paper, we propose a simple but effective novel Local Interpretable model-agnostic Visualizations and Explanations (LIVE) algorithm through combining decision tree and deep learning models. The proposed LIVE algorithm is validated through 4 different data sets with different types of deep learning models including convolutional neural network (CNN) and long short-term memory models (LSTM). Our experiment results show that 1) LIVE algorithm could generally consistently help improve the deep learning models by 0.02 0.03 in AUC; 2) Our LIVE algorithm is a weakly supervised localization tool for region of interest identification; 3) Our LIVE algorithm could be applied to different deep learning models and different types of data.