作者: Cafer Caferov , Barış Kaya , Ryan O’Donnell , AC Cem Say , None
DOI: 10.1007/978-3-662-48054-0_16
关键词:
摘要: Let p be an unknown probability distribution on \([n] := \{1, 2, \dots n\}\) that we can access via two kinds of queries: A SAMP query takes no input and returns \(x \in [n]\) with p[x]; a PMF as the value p[x]. We consider task estimating entropy to within \(\pm \varDelta \) (with high probability). For usual Shannon H(p), show \(\varOmega (\log ^2 n/\varDelta ^2)\) queries are necessary, matching recent upper bound Canonne Rubinfeld. Renyi \(H_\alpha \)(p), where \(\alpha >1\), \(\varTheta (n^{1-1/\alpha })\) necessary sufficient. This complements work Acharya et al. in \(\mathsf \)-only model showed \(O(n^{1-1/\alpha suffice when is integer, but \(\widetilde{\varOmega }(n)\) noninteger. All our lower bounds also easily extend CDF (given x, return \(\sum _{y \le x}\) p[y]) allowed.