이다빈 연구실

I design algorithms with provable guarantees for optimization and sequential decision-making, drawing on mathematical programming, online learning, and reinforcement learning theory. Two questions run through most of my work: how to make good decisions when data arrive over time or depend on the decisions themselves, and how to respect constraints, whether they encode safety, budgets, or combinatorial structure, without sacrificing efficiency. My current projects span online learning and bandits (logistic, neural, and contextual queueing bandits; constrained online convex optimization), reinforcement learning theory (constrained MDPs with adversarial losses, safe RL, and average-reward RL with function approximation), and structured continuous optimization (first-order and projection-free methods for bilevel and constrained problems, bilevel polynomial optimization, and stochastic optimization under decision-dependent distributions). I also keep a long-standing interest in stochastic programming and in the polyhedral and combinatorial theory of integer programming, from cutting planes to clutters. On the applied side, my training in industrial engineering keeps me close to decision problems in production, logistics, and healthcare, where combinatorial structure and uncertainty meet real operational constraints. I have also worked with semiconductor companies on optimization and machine learning problems in memory design and testing, and I welcome collaborations that bring new problems of this kind.