I am interested in all aspects of machine learning, from fully supervised to weakly supervised to unsupervised learning. I am currently working on: (1) Learning + knowledge. In many domains, it may not be easy to obtain a large amount of training data, but we may have had a great deal of knowledge accumulated over a long time. Exploiting knowledge could lead to effective learning with limited data; (2) Explainable AI. In many scenarios , only outputting predictions or even predictions plus statistical confidence may not be sufficient. We need to show the rationale underlying the predictions.