Principles of Coordination in Multi-agent Systems

RiSE hosted a talk by Yoram Moses on April 30, 2015.

ABSTRACT

While distributed and multi-agent systems vary significantly in many parameters including their modes of communication, timing, and topology guarantees, they share an essential locality property: actions at a given site depend on the locally available information, which typically consists of a partial description of the overall state of the system. The first part of this talk will consider three principles underlying coordination in multi-agent systems. The most basic principle has to do with what agents need to know when they perform a given action. It gives rise to two other principles, that capture the information requirements (what agents need to know about other agents) underlying sequential and simultaneous coordination. These principles are universal in the sense that they apply equally well to a set of servers in a datacenter, a set of robots playing soccer, nodes of a wireless sensor network, or ants in an ant colony. The second part of the talk will use the observations made in the first part to capture the role of time and timing information in coordinating actions in systems with clocks. While message chains play a dominant role in asynchronous systems, a corresponding notion is established for systems with clocks and time bounds on message delivery. The talk will not assume prior familairity with the subject. The second part is based on joint work with Ido Ben Zvi.