The goal of this project is developing methods to merge Machine Learning (ML) with physics-based models to create control algorithms can significantly enhance the resiliency of Cyber-Physical Systems (CPS). The approach will combine recent results in ML with control theory via constrained optimization to create novel systems and methods for protecting CPS from malicious cyber intruders via detection and prevention strategies. Specifically, the research project focuses on (1) refining the offline/online training and execution algorithms of ML models through physics-based constrained optimization, (2) developing secure estimation and control algorithms that are significantly more resilient to cyber-attacks than the state-of-the-art counterparts, and (3) improving the distributed resiliency for networked systems supporting multi-agent autonomous systems.