Focuses on optimization and control for trustworthy AI. Builds verification and reachability methods via convex/semidefinite programming, learns Lyapunov and barrier certificates, and develops bilevel and time‑varying optimization to certify safety and performance of neural network–based controllers.