Secure Networked Inference with Unreliable Data Sources

219 Pages·2018·5.69 MB·English

by  Aditya Vempaty

Aditya Vempaty · Bhavya Kailkhura 
Pramod K. Varshney
Secure
Networked
Inference with
Unreliable Data
Sources
Secure Networked Inference with Unreliable Data
Sources
Aditya Vempaty Bhavya Kailkhura
(cid:129)
Pramod K. Varshney
Secure Networked Inference
with Unreliable Data Sources
123
AdityaVempaty PramodK.Varshney
IBMResearch—ThomasJ.Watson Department ofElectrical Engineeringand
Research Computer Science
Yorktown Heights, NY,USA Syracuse University
Syracuse,NY, USA
Bhavya Kailkhura
Department ofComputing Applications and
Research
Lawrence LivermoreNational Laboratory
Livermore, CA,USA
ISBN978-981-13-2311-9 ISBN978-981-13-2312-6 (eBook)
https://doi.org/10.1007/978-981-13-2312-6
LibraryofCongressControlNumber:2018952900
©SpringerNatureSingaporePteLtd.2018
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpart
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
orinformationstorageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilar
methodologynowknownorhereafterdeveloped.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publicationdoesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfrom
therelevantprotectivelawsandregulationsandthereforefreeforgeneraluse.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authorsortheeditorsgiveawarranty,expressorimplied,withrespecttothematerialcontainedhereinor
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictionalclaimsinpublishedmapsandinstitutionalaffiliations.
ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSingaporePteLtd.
Theregisteredcompanyaddressis:152BeachRoad,#21-01/04GatewayEast,Singapore189721,
Singapore
To our families
My parents Anil and Radha
My wife Swetha
Aditya Vempaty
My parents Umesh and Anu
My brother Lakshya
Bhavya Kailkhura
My wife Anju
Pramod K. Varshney
Preface
With an explosion in the number of connected devices and the emergence of big
and dirty data era, new distributed learning solutions are needed to tackle the
problemofinferencewithcorrupteddata.Theaimofthisbookistopresenttheory
and algorithms for secure networked inference in the presence of unreliable data
sources.Morespecifically,wepresentfundamentallimitsofnetworkedinferencein
the presence of Byzantine data (malicious data sources) and discuss robust miti-
gation strategies to ensure reliable performance for several practical network
architectures. In particular, the inference (or learning) process can be detection,
estimation, or classification, and the network architecture of the system can be
parallel, hierarchical, or fully decentralized (peer to peer).
Thisbookisaresultofouractiveresearchoverthepastdecade,wherewehave
been working on the generalization and modernization of the classical “Byzantine
generals problem” in the context of distributed inference to different network
topologies. Over the last three decades, the research community has extensively
studiedtheimpactofimperfecttransmissionchannelsorsensorfaultsondistributed
inferencesystems.However, unreliable(Byzantine) data modelsconsideredinthis
book are philosophically different from the imperfect channels or faulty sensor
cases. Byzantines are intentional and intelligent and therefore can strategically
optimize their actions over the data corruption parameters. Thus, in contrast to
channel-awareinference,boththenetworkandByzantinescanoptimizetheirutility
by choosing their actions based on the knowledge of their opponent’s behavior,
leading to a game between the network and Byzantines. Study of these practically
motivated scenarios in the presence of Byzantines is of utmost importance and is
missing from the channel-aware and fault-tolerant inference literature.
Thisbookprovidesathoroughintroductiontovariousaspectsofrecentadvances
in secure distributed inference in the presence of Byzantines for various network
topologies. Researchers, engineers, and graduate students who are interested in
statisticalinference,wirelessnetworks,wirelesssensornetworks,networksecurity,
data/information fusion, distributed surveillance, or distributed signal processing
willbenefitfromthisbook.Afterreadingthisbook,theyareexpectedtounderstand
theprinciplesbehinddistributedinferenceespeciallyinthepresenceofByzantines,
vii
viii Preface
learnadvancedoptimizationtechniquestosolveinferenceproblemsinthepresence
of practical constraints such as imperfect data, and get insights on inference in ad
hoc networks through practical examples. This book will lay a foundation for
readers to solve real-world networked inference problems. It will be valuable to
students,researchers,andpracticingengineersatuniversities,governmentresearch
laboratories, and companies who are interested in developing robust and secure
inference solutions for real-world ad hoc inference networks. The covered theory
will be useful for a wide variety of applications such as military surveillance,
distributed spectrum sensing (DSS), source localization and tracking, traffic and
environment monitoring, IoT, and crowdsourcing.
Readers should have some background knowledge of probability theory, sta-
tistical signal processing, and optimization. Familiarity with distributed detection
and data fusion (e.g., see P. K. Varshney, Distributed Detection and Data Fusion.
New York: Springer-Verlag, 1997.) will be helpful for some topics, but is not
required to understand the majority of the content.
Yorktown Heights, USA Aditya Vempaty
Livermore, USA Bhavya Kailkhura
Syracuse, USA Pramod K. Varshney
Acknowledgements
This book owes a considerable debt of gratitude to those who have influenced our
work and our thinking over the years.
We are grateful to Hao Chen who initiated this research direction in Sensor
Fusion Laboratory at Syracuse University. The material presented in Sect. 6.2 was
originally conceived by Lav Varshney. We would like to thank Keshav Agrawal,
Priyank Anand, Swastik Brahma, Berkan Dulek, Mukul Gagrani, Yunghsiang S.
Han,WaelHashlamoun,SatishIyengar,QunweiLi,SidNadendla,OnurOzdemir,
Ankit Rawat, Priyadip Ray, and Pranay Sharma for actively contributing to the
research results presented in this book. Kush Varshney and Lav Varshney have
provided critical comments and suggestions during this entire research.
We are also extremely blessed to have supportive families who inspired and
encouraged us to complete this book.
Aditya Vempaty would like to thank his family—parents Anil Kumar Vempaty
and Radha Vempaty, and wife Swetha Stotra Bhashyam—for their unconditional
love, continued support, and extreme patience. Their presence has been a great
mental support and is greatly appreciated.
BhavyaKailkhurawouldliketothankhisparents,grandparents,brother,andthe
restofthefamilyfortheirunconditionallove,support,andsacrificesthroughouthis
life.
PramodK.Varshneyisextremelyblessedtohaveanextremelyupbeatwifewho
inspiredhimthroughouthiscareer.Hededicatestheresearchthatledtothisbookto
his life partner for 40 years, Anju Varshney. She faced severe adversities but
tackled them with a positive attitude. Her motto has been Life Happens Keep
ix
x Acknowledgements
Smiling. His children Lav, Kush, Sonia, and Nina have provided constant
encouragement.HisgrandchildrenArya,Ribhu,andKaushaliaresimplybundlesof
joy and have made his life much happier and cheerful.
July 2018 Aditya Vempaty
Bhavya Kailkhura
Pramod K. Varshney
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Distributed Inference Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation: Byzantine Generals’ Problem . . . . . . . . . . . . . . . . . . 2
1.3 Distributed Inference in the Presence of Byzantines:
Two Viewpoints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Analysis from the Attacker’s Perspective . . . . . . . . . . . . . 3
1.3.2 Analysis from the Network Designer’s Perspective . . . . . . 4
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Foundational Concepts of Inference. . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Binary Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 M-Ary Hypothesis Testing. . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 Asymptotic Performance Metrics . . . . . . . . . . . . . . . . . . . 10
2.2 Distributed Inference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Network Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Practical Concerns. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Taxonomy of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Fundamental Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.2 Mitigation Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Distributed Detection with Unreliable Data Sources . . . . . . . . . . . . . 17
3.1 Distributed Bayesian Detection with Byzantines: Parallel
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.2 Analysis from Attacker’s Perspective . . . . . . . . . . . . . . . . 19
3.1.3 Analysis from Network Designer’s Perspective . . . . . . . . . 28
xi