{"id":"arxiv/2602.01825","url":"https://arxiv.org/abs/2602.01825","tags":["stat.ME","cs.LG","math.OC","stat.ML"],"score":0,"title":"Learning Sequential Decisions from Multiple Sources via Group-Robust Markov Decision Processes","domain":"stat.ME","source":"arxiv","authors":["Mingyuan Xu","Zongqi Xia","Tianxi Cai","Doudou Zhou","Nian Si"],"reviews":0,"subject":"stat.ME","summary":"We often collect data from multiple sites (e.g., hospitals) that share common structure but also exhibit heterogeneity. This paper aims to learn robust sequential decision-making policies from such offline, multi-site datasets. To model cross-site uncertainty, we study distributionally robust MDPs with a group-linear structure: all sites share a common feature map, and both the transition kernels and expected reward functions are linear in these shared features. We introduce feature-wise (d-rectangular) uncertainty sets, which preserve tractable robust Bellman recursions while maintaining…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-02-02","reactions":{}}