IBM Researchers Explain Machine Learning Models By Exploring What Isn't There