Robust Design Discovery and Exploration in Bayesian Optimization

Explore the cutting-edge advancements in robust design discovery and Bayesian optimization through this Google TechTalk presented by Ilija Bogunovic. Delve into the challenges of data-driven decision-making in various fields, including biological design, causal discovery, material production, and physical sciences. Learn about adaptive algorithms and sampling strategies that enable efficient and robust learning in data collection processes. Discover how to quantify uncertainty in optimization objectives, develop robust designs, and create decision-making methods that withstand input perturbations, data shifts, and adversarial attacks. Examine the limitations of existing Bayesian optimization and bandit approaches, and understand novel algorithms that overcome these challenges while maintaining robustness and data efficiency. Gain insights into the practical applications of these algorithms through real-world datasets and popular benchmarks. The talk covers key topics such as Gaussian Process Learning, Confidence Bounds, Robust Bayesian Optimization, Group Robustness, Robust Design Discovery, Model-based Reinforcement Learning, Adversarial Corruptions, and Robust Gaussian Process Phased Elimination.