Please scroll down for article-it is on subsequent pagesINFORMS is the largest professional society in the world for professionals in the fields of operations research, management science, and analytics. For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org Correlation Decay Method for Decision, Optimization, and Inference in Large-Scale Networks
David GamarnikMassachusetts Institute of Technology, Cambridge, Massachusetts 02140, gamarnik@mit.edu Abstract Many inference and decision problems on networks involve solving hard algorithmic problems on graphs. Even when the underlying problem admits a polynomial time algorithm, such an algorithm can be impractical simply because of the sheer size of the underlying network, as many technological, communication, and natural networks may easily be of the order of millions or even billions of nodes. Thus one often has to resort to faster distributed (local) algorithms, which can be run in parallel on the underlying networks, providing the required time and computation effort scales. The correlation decay property, which can be established for a broad class of networks exhibiting random topology, random costs, or randomized decisions, is a tool that provides a rigorous justification for the validity of such local algorithms. In this tutorial we illustrate algorithmic methods based on the correlation decay property on several examples. Starting with the PageRank algorithm for search and navigation on the Internet, we will proceed to applications of the correlation decay method to problems in wireless communication, combinatorial optimization problems on graphs, and, finally, to the problem of network learning-a problem in the machine learning field.