Inference_for_Graphs_and_Networks.pdf

# In the minimax problem the goal is to find procedures

• 99

This preview shows pages 39–41. Sign up to view the full content.

In the minimax problem, the goal is to find procedures that would min- imize ESADD ( T ) and SADD ( T ) subject to E T γ , i.e., in the class C γ . Copyright © 2014. Imperial College Press. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law. EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 2/16/2016 3:37 AM via CGC-GROUP OF COLLEGES (GHARUAN) AN: 779681 ; Heard, Nicholas, Adams, Niall M..; Data Analysis for Network Cyber-security Account: ns224671

This preview has intentionally blurred sections. Sign up to view the full version.

40 A. G. Tartakovsky Note that ideally one would prefer to find a procedure that would min- imize E ν ( T ν | T > ν ) for all ν 0, when the frequency of false alarms is kept at a desired level. However, no such uniformly optimal procedure exists in the class C γ , so we have to resort to minimax or other settings. In the i.i.d. setting, Moustakides (1986) showed that Page’s (1954) CUSUM procedure is strictly minimax with respect to Lorden’s measure of delay to detection ESADD given by (2.2). The CUSUM procedure is based on a maximum likelihood principle: max k 0 Λ k n , which leads (after taking logarithms) to the detection statistic W n = (0 , W n 1 + Z n ) + , n 1 , W 0 = 0 , (2.4) where, as usual, x + = max(0 , x ) denotes a positive part of x , W n = log(max k 0 Λ k n ) and Z n = log L n is the log-likelihood ratio (LLR) for the n th observation. The procedure stops and declares that the change has occurred at T CS ( h ) = min { n 1: W n h } , min { } = , (2.5) i.e., as soon as the detection statistic W n exceeds h , a positive detection threshold preset so as to achieve the desired level of false alarms γ > 1. This can be done by choosing h = h γ = log γ since E T CS ( h ) e h for any h > 0 (Lorden’s, 1971). This choice of the detection threshold guarantees T CS ( h ) C γ , but this lower bound is extremely conservative. For relatively large values of γ a connection of h and γ is given in Tartakovsky (2005): E T CS ( h ) v 1 e h as h → ∞ , where 0 < v < 1 is a computable constant depending on the model. Thus, setting h = log( ) guarantees E T CS ( h ) γ . Hereafter x b y b means that lim b →∞ ( x b /y b ) = 1, which can also be written as x b = y b (1 + o (1)), where o (1) 0 as b → ∞ . A competitive detection procedure, the so-called Shiryaev–Roberts (SR) procedure, is based on the averaging of the likelihood ratio over the uniform prior distribution rather than maximizing it over the unknown changepoint. Specifically, define R n = n k =1 n i = k L i , T SR ( A ) = min { n 1: R n A } , (2.6) where A is a positive threshold that controls the FAR. A connection between A and the ARL2FA E T SR ( A ) is given in Pollak (1987): E T SR ( A ) A for every A > 0 and E T SR ( A ) ζ 1 A as A → ∞ , where the constant 0 < ζ < 1 is subject of renewal theory. Hence, taking A γ = γζ yields ARL2FA ( T SR ) γ for suﬃciently large γ .
This is the end of the preview. Sign up to access the rest of the document.
• Spring '12
• Kushal Kanwar
• Graph Theory, Statistical hypothesis testing, Imperial College Press, applicable copyright law

{[ snackBarMessage ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern