This course provides a rigorous introduction to the security and robustness of deep neural network models, covering both threat models and their corresponding countermeasures. Topics progress from foundational attack paradigms — including adversarial examples and backdoor Trojan attacks — through to system-level threats arising in distributed settings. The federated learning framework is examined in depth, with dedicated coverage of data heterogeneity, Byzantine robustness, and privacy-preserving aggregation. Differential privacy is introduced from first principles and developed through its integration into machine learning pipelines and federated systems. Throughout the course, threat models and defense mechanisms are implemented hands-on in Python using PyTorch. A course project requires students to reproduce and critically analyze a selected publication from the recent literature.