Explore AI-guided nonlinear optimization techniques for solving complex real-world problems in this informative conference talk presented by Yuandong Tian at IPAM's Artificial Intelligence and Discrete Optimization Workshop. Delve into two recent works addressing challenges in efficiently solving optimization problems with highly nonconvex objectives and slow or expensive evaluation processes. Learn about SurCo, a method that leverages neural networks to learn surrogate linear costs for combinatorial constrained problems, and its applications in embedding table sharding and inverse photonics design. Discover CZP, an approach utilizing Transformers to learn analytical parametric forms of frequency responses in linear PDEs, and its effectiveness in antenna design optimization. Gain insights into how these AI-guided techniques can accelerate optimization processes, make use of previously solved instances, and leverage existing solvers to tackle complex real-world optimization challenges.