On Feasible Set Estimation with Bayesian Active Learning

Watch this public lecture exploring Bayesian active learning approaches for feasible set estimation, delivered by Professor Clémentine Prieur from Université Grenoble Alpes at the Isaac Newton Institute for Mathematical Sciences. Discover how Bayesian methods can be strategically applied to efficiently identify and estimate feasible regions in complex parameter spaces through active learning techniques. Learn about the theoretical foundations of feasible set estimation problems and understand how active learning strategies can optimize the selection of evaluation points to maximize information gain while minimizing computational costs. Explore practical applications where determining feasible parameter regions is crucial, such as engineering design optimization, safety analysis, and reliability assessment. Examine the mathematical framework underlying Bayesian active learning for set estimation, including uncertainty quantification, acquisition functions, and convergence properties. Understand how these methods balance exploration and exploitation to efficiently map boundaries between feasible and infeasible regions. The lecture covers both theoretical developments and practical implementation considerations, making it valuable for researchers and practitioners working in optimization, uncertainty quantification, machine learning, and applied mathematics who need to solve feasible set estimation problems efficiently.