{"id":"arxiv/2601.22005","url":"https://arxiv.org/abs/2601.22005","tags":["quant-ph","math.ST","stat.ML"],"score":0,"title":"Hierarchy of discriminative power and complexity in learning quantum ensembles","domain":"quant-ph","source":"arxiv","authors":["Jian Yao","Pengtao Li","Xiaohui Chen","Quntao Zhuang"],"reviews":0,"subject":"quant-ph","summary":"Distance metrics are central to machine learning, yet distances between ensembles of quantum states remain poorly understood due to fundamental quantum measurement constraints. We introduce a hierarchy of integral probability metrics, termed MMD-$k$, which generalizes the maximum mean discrepancy to quantum ensembles and exhibits a strict trade-off between discriminative power and statistical efficiency as the moment order $k$ increases. For pure-state ensembles of size $N$, estimating MMD-$k$ with arbitrary measurement schemes requires $Θ(N^{1-1/k})$ samples for constant $k$. At the same…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-01-29","reactions":{}}