A prompt is routed to either a small model or a large model. Each route produces a feedback symbol revealed to the user and a hidden reward. Only revealed symbols guide learning and future routing, with the goal of maximizing cumulative reward.
Efficient contextual partial monitoring with general function approximation
Kapilan Balagopalan, Hao Qin, Kwang-Sung Jun, Chicheng Zhang
arXiv preprint, 2026
Paper
Choose 3 of 6 items based on context. Observe each chosen item's reward and learn to maximize total reward. Unchosen rewards stay unknown.
Optimal and Efficient Contextual Combinatorial Semi-bandits with General Function Approximation
Hao Qin, Chicheng Zhang
Conference on Neural Information Processing Systems (NeurIPS) 2026
Paper
DOEC
Taming the Monster Every Context: Complexity Measure and Unified Framework for Offline-Oracle Efficient Contextual Bandits
Hao Qin, Chicheng Zhang
Conference on Learning Theory (COLT) 2026
Paper | Slides | Poster
UNIMODAL
Physics-Informed Parametric Bandits for Beam Alignment in mmWave Communications
Hao Qin*, Thang Duong*, Ming F. Li, Chicheng Zhang
International Symposium on Modeling and Optimization in Mobile, Ad hoc, and Wireless Networks (WiOpt) 2026
Paper | Slides
EXP-KL-MS
Achieving Adaptivity and Optimality for Multi-Armed Bandits using Exponential-Kullback Leibler Maillard Sampling
Hao Qin, Kwang-Sung Jun, Chicheng Zhang
Transactions on Machine Learning Research (TMLR) 2026 — Featured Certification
Paper
KL-MS
Kullback-Leibler Maillard Sampling for Multi-armed Bandits with Bounded Rewards
Hao Qin, Kwang-Sung Jun, Chicheng Zhang
Conference on Neural Information Processing Systems (NeurIPS) 2023
Paper | Code