Asynchronous distributed decision-making
Abstract
Classical (centralized) theories of decision making and computation deal with the situation in which a single decision maker (man or machine) possesses (or collects) all available information related to a certain system and has to perform some computation and/or make a decision so as to achieve a certain objective. We are concerned with systems were the decision are shared among independent decision sites. Lack of centralized control promises to overcome communication and/or computation bottlenecks that might inhibit computational efficiency. Asynchronism is introduced as an alternative allowing for the independent operation of the computing elements. The possibility of a delayed information can generate results whose value can never be determined. An attempt is made to derive, for some relaxed instances of asynchronous iterations, generic forms of the equations that describe the dynamics of asynchronous relaxation, and based on these relaxed models we will attempt to describe, and, when this is possible, to predict the dynamic response of iterative schemes in the presence of delays. Motivated by the concepts and ideas of asynchronous computing we study optimal decision making problems exhibiting, to a certain extent, decentralized characteristics, namely, they can be perceived as a collection of sub-systems, each one characterized by its local properties and dynamics, joined together by the need to accomplish a certain common task in order to form the overall decision making problem. There are two main characteristics of such systems that make them ideally suited for the framework of asynchronous computing. First of all, an overall objective describing the collective behavior of all the sub-systems need not exist, but even if it does, it does not have to be known by all of the sub-systems. As long as the effect of the interactions can be perceived by the partial decision makers, the complex system can still be driven to optimality. The second characteristic, but of equal importance, is that once an optimal decision has been made, based on some local in space and time information, by a certain decision maker, this decision has to be implemented with no further delay. Algorithms for distributed decision making based on the ideas of asymptotic agreement and overlapping computation are presented.
Degree
Ph.D.
Advisors
Reklaitis, Purdue University.
Subject Area
Chemical engineering|Systems design
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.