Resilient Consensus of Multi-Agent Networks in the Presence of Malicious Attacks

Wednesday, March 15, 2017, 11:00 am - 12:00 pm PDTiCal
ISI, Conference Room 689
This event is open to the public.
CPS Seminar
Seyed Mehran Dibaji

The focus of the talk is on resilient consensus problem where the aim is to reach agreement on specific variables that the agents have, but certain agents in the network anonymously try to mislead others. Such malicious agents do not follow the predefined local interaction rules and might prevent the normal agents to coordinate properly. The malicious agents are omniscient in the sense that they have global information regarding the network. This type of problems have been studied in the area of distributed algorithms in computer science since 1980s. However, our approach is based on the viewpoint of multi-agent control, which has provided new motivations and problem settings more realistic for UAV type applications. We have incorporated the dynamics of the agents and have also found conditions on the network structure of agents interactions. To enhance the security level and to avoid being influenced by malicious agents, we have introduced a simple rule where each normal agent ignores some of its neighbors which take the most deviated values. Such algorithms are known as Mean Subsequence Reduced (MSR) algorithms in computer science and more recently in control. In particular, we concentrate on resilient consensus of first-order, second-order and quantized agents under asynchronous update times and information delays.

« Return to Upcoming Events