Prof. Hanzhang Qin
National University of Singapore
Talk: Toward Agentic Decision-Making in Operations: Bandits with Generative Signals
Abstract: Recent advances in large language models (LLMs) are enabling a new paradigm of agentic decision-making, where AI systems can autonomously generate, evaluate, and refine decisions in complex operational environments. Frameworks such as ORLM, OptiMUS and LEAN-LLM-OPT illustrate this vision by integrating LLMs into end-to-end optimization pipelines. In this talk, I focus on two foundational challenges underlying this paradigm: how to efficiently explore decision spaces shaped by generative models, and how to safely incorporate learned signals into decision-making. First, I present a structured bandit framework for prompt optimization, where candidate prompts are embedded into a latent space and explored using a hierarchical Thompson sampling algorithm. By combining soft clustering with Bayesian learning, the method efficiently balances global semantic exploration and local refinement, significantly improving sample efficiency under strict evaluation budgets. Second, I introduce a learning-augmented bandit framework in which each action yields both a reward and a proxy signal generated by predictive or generative models. We develop a certified control variate approach that adaptively verifies and exploits these proxy signals, ensuring that performance never degrades relative to classical methods while achieving provable reductions in sample complexity.
Biography: Hanzhang Qin is an Assistant Professor at the Department of Industrial Systems Engineering and Management at NUS. He is also an affiliated faculty member at the NUS Institute for Operations Research and Analytics and the NUS AI Institute. Prof. Qin’s research interests span agentic decision-making, applied probability and statistical learning, with applications in supply chain analytics and transportation systems. His research was recognized by several awards, including INFORMS TSL Intelligent Transportation Systems Best Paper Award, APORS Young Researcher Best Paper Award and MIT MathWorks Prize for Outstanding CSE Doctoral Research. Prof. Qin graduated from MIT with a PhD in Computational Science and Engineering and 2 Masters (EECS and Transportation), and received two bachelor degrees in Industrial Engineering and Mathematics from Tsinghua University.Before joining NUS, he spent one year as a postdoctoral scientist in the Supply Chain Optimization Technologies Group of Amazon NYC.
