{"id":"arxiv/2605.01335","url":"https://arxiv.org/abs/2605.01335","tags":["stat.ML","cs.LG","math.ST"],"score":0,"title":"Mean Testing under Truncation beyond Gaussian","domain":"stat.ML","source":"arxiv","authors":["Yuhao Wang","Roberto Imbuzeiro Oliveira","Themis Gouleakis"],"reviews":0,"subject":"stat.ML","summary":"We characterize the fundamental limits of high-dimensional mean testing under arbitrary truncation, where samples are drawn from the conditional distribution $P(\\cdot \\mid S)$ for an unknown truncation set $S$ that may hide up to an $\\varepsilon$-fraction of the probability mass. For distributions with $p$-th directional moments of magnitude at most $ν_{P,p}$, truncation induces a bias of order $O(ν_{P,p}\\varepsilon^{1-1/p})$. This bias creates a sharp information-theoretic detectability floor: when the signal $α$ falls below this threshold, the null and alternative hypotheses are…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-05-02","reactions":{}}