Agents Overcoming Resource Independent Scaling Threats (AORIST)
Abstract
This project uses abstract simulation models of resource allocation and mathematical techniques inspired by statistical physics to study the nonlinear emergent dynamics of distributed decentralized resource allocation. Our techniques seek to characterize the dynamics that may be anticipated in real systems, to predict pathological dynamics such as peaks in required computational effort and catastrophic breakdown in performance, and to develop control methods based on this understanding. Our general approach begins with a set of abstract Resource Allocation Games (RAG). These games are derived from the Minority Game, a simple model of competition for scarce re-sources that captures essential features of interactions among agents that are heterogeneous, autonomous, boundedly rational, adaptable, parallel, co-situated, and experienced. Our research explores and generally confirms two hypotheses concerning the dynamics of resource allocation. The Generality Hypothesis asserts that a generic RAG exhibits dynamics that are intrinsic to resource allocation in-dependent of mechanism. The Specificity Hypothesis asserts that a RAG can be developed to resemble a specific re-source allocation mechanism and study its (idiosyncratic) dynamics.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2004
- Accession Number
- ADA428683
Entities
People
- H. Van Dyke Parunak
- John A. Sauter
- Robert Savit
- Sven Brueckner