DocumentCode :
2048418
Title :
Communication Complexity under Product and Nonproduct Distributions
Author :
Sherstov, Alexander A.
Author_Institution :
Dept. of Comput. Sci., Univ. of Texas at Austin, Austin, TX
fYear :
2008
fDate :
23-26 June 2008
Firstpage :
64
Lastpage :
70
Abstract :
We solve an open problem in communication complexity posed by Kushilevitz and Nisan (1997). Let Repsiv(f) and Dmu epsiv(f) denote the randomized and mu-distributional communication complexities off, respectively (e a small constant). Yao´s well-known minimax principle states that Repsiv(f) = maxmu{Dmu epsiv(f)}. Kushilevitz and Nisan (1997) ask whether this equality is approximately preserved if the maximization is taken over product distributions only, rather than all distributions mu. We give a strong negative answer to this question. Specifically, we prove the existence of a function f : {0,1}n X {0,1}n rarr {0, 1}for which Repsiv(f) = Omega(n) but maxmuproduct {Dmu epsiv(f)} = 0(1).
Keywords :
communication complexity; minimax techniques; random processes; minimax principle state; mu-distributional communication complexity; randomized communication complexity; Complexity theory; Computational complexity; Costs; Distributed computing; Error probability; Game theory; Minimax techniques; Probability distribution; Protocols; USA Councils; Randomized/distributional communication complexity; Yao´s Minimax Principle; product/nonproduct distributions;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Computational Complexity, 2008. CCC '08. 23rd Annual IEEE Conference on
Conference_Location :
College Park, MD
ISSN :
1093-0159
Print_ISBN :
978-0-7695-3169-4
Type :
conf
DOI :
10.1109/CCC.2008.10
Filename :
4558810
Link To Document :
بازگشت