{"id":"arxiv/2605.08866","url":"https://arxiv.org/abs/2605.08866","tags":["stat.ML","cs.LG","math.OC"],"score":0,"title":"Tight Generalization Bounds for Noiseless Inverse Optimization","domain":"stat.ML","source":"arxiv","authors":["Pouria Fatemi","Hoomaan Maskan","Suvrit Sra","Peyman Mohajerin Esfahani"],"reviews":0,"subject":"stat.ML","summary":"Inverse optimization (IO) seeks to infer the parameters of a decision-maker's objective from observed context--action data. We study noiseless IO, where demonstrations are generated by a ground-truth objective. We provide a high-probability ${O}(\\frac{d}{T})$ generalization bound for the induced action set, where $d$ is the number of unknown parameters and $T$ is the size of the training dataset. We strengthen these guarantees under additional conditions that ensure uniqueness of the chosen action, bringing our IO guarantees in line with best-arm identification results in the bandit…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-05-09","reactions":{}}