Protecting Models and Training Data: Extraction, Membership Inference and Defenses