He is working primarily on robustness and interpretability of machine learning algorithms, along with models that reason with text and structure for natural language processing.