Bi Level Optimization

Bi-level optimization (BLO) is a mathematical framework for solving hierarchical optimization problems, where an outer problem's solution depends on the solution of an inner problem. Current research focuses on developing efficient and stable algorithms, particularly gradient-based methods, to address challenges like memory limitations in large-scale applications and the instability arising from non-convexity in both levels. BLO finds applications across diverse fields, including machine learning (e.g., hyperparameter optimization, neural architecture search, federated learning), and control systems, offering improved model performance and efficiency. The development of more robust and scalable BLO algorithms is crucial for advancing these applications.