Machine Learning is rapidly becoming ubiquitous in computation, from deep learning for images, speech and language to large-scale data mining and decision support. But some major challenges remain including: 1) how to cope with sparse expert answers (labels) to train accurate models, 2) how to explain the learned behaviors, 3) how to combine knowledge and constraints with data, especially when the latter is scarce. The presentation will introduce proactive learning from multiple sources, transfer/multi-task learning, and address issues in application of these methods to different areas, such as natural language processing and computational biology