Adaptive Control Based on the Use of Collective Information from Multiple Models

Yale University, New Haven CT

Investigators

Abstract

Intellectual Merit: The objective of the proposal is to develop mathematical theory of decision making under uncertainty based on multiple models. How the different views should be combined to arrive at decisions rapidly and accurately at any instant is addressed. Past efforts were based on having a sufficiently large number of experts and basing the current decision on the one proven most successful in the short term. In contrast to such approaches, the proposal deals with novel procedures, in which the decisions of all the experts are weighted to choose the action at any instant. The weights, in turn, depend upon the recent short term performance of all the participants. Simulation studies have indicated that the new approach results in significantly faster and more accurate decisions, and that the latter are more robust even under rapidly changing environments. Broader Impacts: The methodology investigated in the proposal will be applicable to a very large class of problems. In particular, the new technique will find application in Cyber-Physical systems in which vast amounts of data are collected by numerous sensors, and networking problems in which distributed decision makers have to make critical choices in real time. The results obtained from the investigations will be widely disseminated by the PI and his co-workers in national and international conference, and in specialized workshops held at Yale. More importantly, undergraduates will be trained by the PI, both during the academic year and during the summers, in the mathematical foundations of the approach as well as in their application to practical problems.