Distributed Inference in the Presence of Byzantines Pramod K. Varshney Distinguished Professor, EECS Director of CASE: Center for Advanced Systems and Engineering Syracuse University E-mail: varshney@syr.edu in collaboration with K. Agrawal, P. Anand, A. Rawat B. Kailkhura, V. S. S. Nadendla, A. Vempaty S. K. Brahma , H. Chen , Y.S. Han, O. Ozdemir
Distinguished Professor, EECS Director of CASE: Center for Advanced Systems and Engineering Syracuse University E-mail: varshney@syr.edu in collaboration with K. Agrawal, P. Anand, A. Rawat B. Kailkhura, V. S. S. Nadendla, A. Vempaty S. K. Brahma , H. Chen , Y. S. Han, O. Ozdemir
Current Topics of Interest Distributed Inference Detection, Estimation, Classification, Tracking Fusion for Heterogeneous Sensor Networks e Modeling(dependent sensors using copula theory) Sensor Management (Traditional/ Game-theoretic designs compressed Inference Stochastic resonance pplications Cognitive Radio Networks Ecological monitoring e Security for Spectrum Sensing Acoustic monitoring of wildlife pectrum Auctions in national forest reserves o Reliable crowdsourcing o Medical Image Processing
Distributed Inference Detection, Estimation, Classification, Tracking Fusion for Heterogeneous Sensor Networks Modeling (Dependent sensors using copula theory) Sensor Management (Traditional/Game-theoretic designs) Compressed Inference Stochastic Resonance Cognitive Radio Networks Security for Spectrum Sensing Spectrum Auctions Reliable Crowdsourcing Ecological monitoring Acoustic monitoring of wildlife in national forest reserves Medical Image Processing
Outline e Distributed inference and data fusion ● Byzantine attacks stributed Inference with Byzantines Ongoing research and Future Work
Distributed Inference and Data Fusion Byzantine Attacks Distributed Inference with Byzantines Ongoing Research and Future Work
Distributed Inference in practice E UNI ED A usIla HEALTH MONTORING AVIATION BODY SENSOR NETWORKS) DIAGNOSTICS ECOLOGICAL DEFENSE MONITORING (DISTRIBUTED RADAR, UGS) 单空::
Different sensors, Diverse Information VIDEO THZ IMAGING ACOUSTIC SEISMIC THROUGH-THE-WALL
Multi-sensor Inference: Information usion e Typical decision making processes involve combining information from various sources Designing an automatic system to do this is a challenging tas k o Many benetits from such a system Common source overage Multiple sensors > Robust system Fusion center nformation Diversity
Typical decision making processes involve combining information from various sources Designing an automatic system to do this is a challenging task Many benefits from such a system Common source Multiple sensors Fusion center Coverage Robust system Information Diversity
Six Blind men and an elephant It was six men of indostan To learning much inclined Who went to see the elephant (Though all of them were blind) That each by observation Might satisfy his mind The First approached the elephant, And happening to fall Against his broad and sturdy side At once began to bawl God bless me! but the elephant Is very like a wall! And so these men of Indostan Disputed loud and lon Each in his own opinion Exceeding stiff and strong Though each was partly in the right And all were in the wrong John Godfrey Saxe
It was six men of Indostan To learning much inclined, Who went to see the Elephant (Though all of them were blind), That each by observation Might satisfy his mind. The First approached the Elephant, And happening to fall Against his broad and sturdy side, At once began to bawl: "God bless me! but the Elephant Is very like a wall!" …… And so these men of Indostan Disputed loud and long, Each in his own opinion Exceeding stiff and strong, Though each was partly in the right, And all were in the wrong! - John Godfrey Saxe
Inference Network Phenomenon S-1 2 S3 SN 2 43 Fusion Center lo Sensors collect raw-observations and transmit processed-observations to the fusion center e Fusion center makes global inferences based on the sensor messages e Inferences: Detection, Estimation. Classification
Inference Network Phenomenon S-1 S-2 S-3 S-N Fusion Center y1 y2 y3 yN u0 u1 u2 u3 uN ... Sensors collect raw-observations and transmit processed-observations to the fusion center. Fusion center makes global inferences based on the sensor messages. Inferences: Detection, Estimation, Classification
Typical Inference Problems and Applications De detection ● Estimation ● Example: Spectrum Example: State Estimation Sensing in Cognitive Radio in Smart Grids etworks Fusion Center Customers Users(SUs) Generator Primary User (PU)
Detection Example: Spectrum Sensing in Cognitive Radio Networks Estimation Example: State Estimation in Smart Grids Primary User (PU) Secondary Users (SUs) Fusion Center . . . . . . . . .
Centralized vs Distributed Inference Phenomenon Phenomenon S-1 S-2 SN S-2 IS-N Decision 1 Decision n usion center Decision 1 Decision n Global Decision Centralized inference e Distributed inference All the sensor signals are assumed Distributed processing to be available in one place for Decision rules, both at the local processing sensors and at the fusion center, Each detector acts independently are based on system wide joint and bases its decision on likelihood optimization ratio test (Lrt)
Centralized Inference All the sensor signals are assumed to be available in one place for processing Each detector acts independently and bases its decision on likelihood ratio test (LRT) Distributed Inference Distributed processing Decision rules, both at the local sensors and at the fusion center, are based on system wide joint optimization Phenomenon S-1 S-2 S-3 S-N Fusion Center y1 y2 y3 yN u0 u1 u2 u3 uN Local Decision 1 Local Decision N Global Decision Phenomenon S-1 S-2 S-3 S-N y1 y2 y3 yN u1 u2 u3 uN Decision 1 . . . . . Decision N . . . . . . . . .