{"id":"arxiv/2610.42109","url":"https://arxiv.org/abs/2610.42109","tags":["stat.ML","cs.LG","math.OC"],"score":0,"title":"Regret-Optimal Wasserstein-Robust Linear Regression","domain":"stat.ML","source":"arxiv","authors":["Sloan Nietert","Daniel Kuhn"],"reviews":0,"subject":"stat.ML","summary":"Distributionally robust optimization (DRO) minimizes worst-case risk under distributional uncertainty, but its decisions can be suboptimal in terms of their excess risk, or regret. We study this gap for multivariate least-squares regression, assuming that the observed source law and unknown target law differ by an adversarial 2-Wasserstein perturbation of size $\\varepsilon$. We first identify the asymptotic complexity of robust decision-making in the $\\varepsilon \\to 0$ limit, characterized by a problem-specific condition number $κ$. Here, a wide spectrum of simple algorithms, including DRO,…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-10-30","reactions":{}}