Adaptive choice of information sources

Abstract

We present a number of learning approaches by which agents can adapt to select information sources that satisfy performance requirements. Performance can be interpreted both in terms of the quality of information provided by the sources, as well as the response time to process information requests. We first present a couple of approaches by which self-motivated agents can learn to choose lightly-loaded resources. The resultant load balancing effect results in increasing throughput for the entire system as well as faster response times for individual agents. We also present an expected utility maximization approach to selecting information sources that are likely to deliver better quality information to different classes of queries.