Hacking Smart Machines with Smarter Ones: How to Extract Meaningful Data from Machine Learning Classifiers