Protocol Design for Decentralized Coordination
Abstract
Protocol Design for Decentralized Coordination Yevgeniy Vorobeychik Assistant Professor, EECS, Vanderbilt University Abstract Large-scale naval missions rely on a complex of loosely-coupled moving parts acting at a high degree of synchrony towards a set of common goals. The emergence of such synchrony of action by multiple decision makers hinges on highly efficient coordination among them under tight time constraints and, often, in adversarial environments. Decentralized coordination of this kind can succeed, or fail, depending on the protocol governing interactions among the decision makers. The goal of the proposed research, therefore, is to address the problem of designing protocols that promote highly efficient decentralized coordination. The proposed approach will involve two interdependent thrusts: first, human subject experiments in decentralized coordination on networks, focusing on the effectiveness of protocols at increasing coordination efficiency, and second, mathematical and computational modeling and simulation to investigate the problem of protocol design at scale. The proposed experimental research thrust will focus on two means of promoting efficient coordination: communication among decision makers, and a reputation system. We will consider two broad classes of communication media: 1) local, where subjects only communicate with their immediate network neighbors, and 2) central communication hub (message board), to which all subjects can post messages which would subsequently be globally visible. We will investigate both free-text (unstructured) and restricted (structured) communication, where in the former there will be a hard constraint on message length, as well as a per-character cost. In addition, we will explore how a simple reputation system, whereby players can score each other, with the aggregate score representing a player’s globally visible reputation, affects the efficacy of coordination. The proposed experimental research will additionally investigate the ability of adversaries embedded in a network to prevent coordination, as well as the effectiveness of communication and reputation systems at identifying such adversaries. The proposed modeling and simulation effort will focus on two subproblems: 1) modeling of individual human behavior based on experimental data, and 2) a simulation-based optimization framework for identifying high-quality coordination protocols. The modeling research will focus on parametric continous-time mathematical representations of human subject behavior which can be calibrated and validated on human subject data. Protocol optimization would then use such validated models, combined with stochastic local search and machine learning, to compute and compactly represent high-quality protocols. 1
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512621
Entities
People
- Yevgeniy Vorobeychik
Organizations
- Office of Naval Research
- United States Navy
- Vanderbilt University