Bilevel Optimization

Bilevel optimization tackles nested optimization problems, aiming to optimize an upper-level objective function whose solution depends on the optimal solution of a lower-level problem. Current research focuses on developing efficient first-order algorithms, often avoiding computationally expensive Hessian computations, and addressing challenges like non-convexity, stochasticity, and constraints in both levels. This framework finds applications across diverse fields, including machine learning (hyperparameter optimization, meta-learning, model training), control systems, and scientific discovery, offering improved efficiency and robustness in solving complex, multi-stage optimization tasks.