TY - GEN
T1 - Resilient Distributed Hypothesis Testing with Time-Varying Network Topology
AU - Wu, Bo
AU - Carr, Steven
AU - Bharadwaj, Suda
AU - Xu, Zhe
AU - Topcu, Ufuk
N1 - Funding Information:
Bo Wu, Steven Carr, Suda Bharadwaj, Zhe Xu, and Ufuk Topcu are with the University of Texas at Austin. emails: {bwu3, stevencarr, suda.b, zhexu, utopcu}@utexas.edu. This work was partly funded by grants AFRL FA9550-19-1-0169, DARPA D19AP00004, Sandia National Lab 801KOB.
Publisher Copyright:
© 2020 AACC.
PY - 2020/7
Y1 - 2020/7
N2 - We study the problem of distributed hypothesis testing, where a team of mobile agents aims to agree on the true hypothesis (out of a finite set of hypotheses) that best explains a sequence of their local and possibly noisy observations. The setting requires team collaborations through a time-varying network topology due to mobility and limited communication range. We also assume that there is an unknown subset of compromised agents that may deliberately share wrong information to undermine the team objective. We propose a distributed algorithm where each agent maintains two sets of beliefs (i.e., probability distributions over hypotheses), namely local and actual beliefs. For each agent at each time step, the local belief is updated based on its local observations. Then the actual belief is updated with its local belief and shared actual beliefs from the other agents within the communication range. We show that the actual belief of each non-adversarial agent converges almost surely to the true hypothesis. Unlike most of the existing literature, we guarantee the convergence without a connectivity constraint of the time-varying network topology.
AB - We study the problem of distributed hypothesis testing, where a team of mobile agents aims to agree on the true hypothesis (out of a finite set of hypotheses) that best explains a sequence of their local and possibly noisy observations. The setting requires team collaborations through a time-varying network topology due to mobility and limited communication range. We also assume that there is an unknown subset of compromised agents that may deliberately share wrong information to undermine the team objective. We propose a distributed algorithm where each agent maintains two sets of beliefs (i.e., probability distributions over hypotheses), namely local and actual beliefs. For each agent at each time step, the local belief is updated based on its local observations. Then the actual belief is updated with its local belief and shared actual beliefs from the other agents within the communication range. We show that the actual belief of each non-adversarial agent converges almost surely to the true hypothesis. Unlike most of the existing literature, we guarantee the convergence without a connectivity constraint of the time-varying network topology.
UR - http://www.scopus.com/inward/record.url?scp=85089580881&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85089580881&partnerID=8YFLogxK
U2 - 10.23919/ACC45564.2020.9148036
DO - 10.23919/ACC45564.2020.9148036
M3 - Conference contribution
AN - SCOPUS:85089580881
T3 - Proceedings of the American Control Conference
SP - 1483
EP - 1488
BT - 2020 American Control Conference, ACC 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2020 American Control Conference, ACC 2020
Y2 - 1 July 2020 through 3 July 2020
ER -