{"id":"arxiv/2605.21860","url":"https://arxiv.org/abs/2605.21860","tags":["math.ST","cs.DS","cs.IT","stat.ML"],"score":0,"title":"Robust Statistical Estimators with Bounded Empirical Sensitivity","domain":"math.ST","source":"arxiv","authors":["Valentio Iverson","Gautam Kamath","Argyris Mouzakis","Adam Smith"],"reviews":0,"subject":"math.ST","summary":"We introduce a new measure of robustness for statistical estimators, which we call \\emph{empirical sensitivity}. An estimator $\\hat θ$ has bounded empirical sensitivity if, with high probability over a dataset $X = (X_1, \\dots, X_n) \\sim \\mathcal{D}^{\\otimes n}$, for any dataset $Y$ obtained by modifying at most $ηn$ points in $X$, we have that $\\hat θ(Y)$ is close to $\\hat θ(X)$. We study bounds on this quantity for the prototypical problem of Gaussian mean estimation. We prove new lower bounds, showing that for any estimator $\\hat μ$ which achieves an optimal $\\ell_2$-error bound of…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-21","reactions":{}}