Complex Systems

Learning Decentralized Goal-based Vector Quantization Download PDF

Piyush Gupta
Electronic mail address:
Department of Electrical and Computer Engineering,
University of Illinois at Urbana-Champaign, Urbana, IL

Vivek S. Borkar
Electronic mail address:
Department of Computer Science and Automation,
Indian Institute of Science, Bangalore, India


This paper considers the following generalization of classical vector quantization: to (vector) quantize or, equivalently, to partition the domain of a given function such that each cell in the partition satisfies a given set of topological constraints. This formulation is called decentralized goal-based vector quantization (DGVQ). The formulation is motivated by the resource allocation mechanism design problem from economics. A learning algorithm is proposed for the problem. Various extensions of the problem, as well as the corresponding modifications in the proposed algorithm, are discussed. Simulation results of the proposed algorithm for DGVQ and its extensions, are given.