Haichen Hu

Haichen Hu is a graduate student in computational science and engineering and civil and environmental engineering whose work bridges statistical learning, reinforcement learning, and optimization for decision-making under uncertainty in operations research. In reinforcement learning, Haichen developed one of the first optimal algorithms with low-frequency oracle calls for policy optimization and offline estimation, substantially reducing computational overhead. More broadly, he unifies sequential decision-making problems—including bandits, model calibration, and active learning—through functional analysis and operator theory, establishing how eigendecay of compact operators governs regret rates. He also established tight excess-risk bounds for black-box predictors in data- and compute-efficient settings, and most recently developed a black-box reduction from nonconvex to online convex optimization for efficient large language model training. As a MathWorks Fellow, Haichen will develop optimization and statistical learning methods to solve combinatorial optimization problems in real operations research applications. His goal is to make black-box-assisted decision-making more effective in operational settings involving sequential predictions and multiple steps.