Using Trust in Distributed Consensus with Adversaries in Sensor and Other Networks Xiangyang Liu, and John S. Baras Institute for Systems Research and Department of Electrical and Computer Engineering University of Maryland, College Park, MD
Outline Introduction Motivation Trust-Aware Consensus Simulations Conclusion
Introduction Cooperation Agent Cooperation Distributed sensor fusion. Goal: all agents reach consensus on ML estimate. [1] Xiao, Lin, Stephen Boyd, and Sanjay Lall. "A scheme for robust distributed sensor fusion based on average consensus." Information Processing in Sensor Networks, IPSN Fourth International Symposium on. IEEE, Distributed Coordination. Goal: all agents reach decision on same direction (location) [2] Jadbabaie, Ali, Jie Lin, and A. Stephen Morse. "Coordination of groups of mobile autonomous agents using nearest neighbor rules." Automatic Control, IEEE Transactions on 48.6 (2003): Agent Without supervisor
Introduction Cooperation Agent Cooperation Agent Link Jam & Noise Injection: [3]Khanafer, Ali, Behrouz Touri, and Tamer Basar. "Consensus in the presence of an adversary." 3rd IFAC Workshop on Distributed Estimation and Control in Networked Systems (NecSys) Malicious Agent Malicious agent: Multiparty secure computation [4] Garay, Juan A., and Rafail Ostrovsky. "Almost-everywhere secure computation." Advances in Cryptology–EUROCRYPT Springer Berlin Heidelberg, Consensus with Byzantine adversaries (System theory) [5] Pasqualetti, Fabio, Antonio Bicchi, and Francesco Bullo. "Consensus computation in unreliable networks: A system theoretic approach." Automatic Control, IEEE Transactions on 57.1 (2012): trust
Motivation Good Node Malicious Node Goal: Detect malicious nodes and isolate them from consensus algorithm.
Trust-Aware Consensus Good Node Malicious Node Trust Evidence Local Trust Decision rules Global Trust Trust Propagation Trust-Aware Consensus Embed trust into consensus
Trust-Aware Consensus Trust Evidence Local Trust Node i’s trust evidence: Clustering-Based Distance-Based Consistency-Based Decision rules:
Trust-Aware Consensus Clustering-Based Distance-Based
Trust-Aware Consensus Consistency-Based message broadcast by node l and heard by node i message broadcast by node j about what it hears from node l
Trust-Aware Consensus Local Trust Global Trust Trust Propagation Malicious Normal Header
Trust-Aware Consensus Trust Evidence Local Trust Decision rules Global Trust Trust Propagation Trust-Aware Consensus Embed trust into consensus
Simulations Adversary outputs constant message. Figure on the left has no trust propagation. Figure on the right has trust propagation.
Simulations Adversary switches randomly between several messages. Figure on the left has no trust propagation. Figure on the right has trust propagation.
Simulations Adversary takes random noise strategy. Figure on the left has no trust propagation. Figure on the right has trust propagation.
Simulations Adversary takes fixed noise strategy. Figure on the left has no trust propagation. Figure on the right has trust propagation.
Simulations Left: adversary takes constant strategy. Right: adversary takes random noise strategy. The communication graph has connectivity 2.
Conclusion Developed trust model with various decision rules based on local evidence in the setting of Byzantine adversaries. Trust-Aware consensus algorithm proposed is flexible and can be extended to incorporate more complicated trust models and decision rules. Simulations show our algorithm can effectively detect malicious strategies even in sparse networks of connectivity, where is the number of adversaries.
Thank you! {xyliu, Questions?