Mitigation of Byzantine Attacks on Distributed Detection Systems Using Audit Bits

Wael Hashlamoun, Swastik Brahma, Pramod K. Varshney

Research output: Contribution to journalArticlepeer-review

19 Scopus citations


This paper considers the problem of distributed detection in the presence of Byzantines who seek to degrade detection performance by falsifying data. This paper proposes a novel mechanism to mitigate Byzantine attacks by partitioning sensors into groups. Local decisions from sensors in each group are sent to the Fusion Center (FC) via multiple paths, which enable the FC to assess (i.e., to audit) the information that reaches it to improve detection performance. We introduce a weighted Kullback-Leibler Divergence (KLD) metric to measure the detection performance of the FC and show that the proposed mechanism is more robust against Byzantine attacks than previously proposed schemes. We prove that, using the proposed mechanism, the FC becomes blind (i.e., no useful information reaches the FC) only if all the nodes in the network are Byzantines. This paper also characterizes optimal Byzantine attacks in the scenario when the FC cannot be made blind. Further, the paper conducts game theoretic analysis to investigate the scenario when both the Byzantines and the FC act strategically against each other and prove the existence of a Nash Equilibrium. Extensive numerical results are provided throughout the paper that provide insights into the proposed distributed detection mechanism.

Original languageEnglish (US)
Pages (from-to)18-32
Number of pages15
JournalIEEE Transactions on Signal and Information Processing over Networks
Issue number1
StatePublished - Mar 2018


  • Byzantine attacks
  • Kullback-Leibler divergence
  • distributed detection
  • game theory

ASJC Scopus subject areas

  • Signal Processing
  • Information Systems
  • Computer Networks and Communications


Dive into the research topics of 'Mitigation of Byzantine Attacks on Distributed Detection Systems Using Audit Bits'. Together they form a unique fingerprint.

Cite this