Exploiting Multi-Agent Interactions for Identifying the Best-Payoff Information Source - Robotics Institute Carnegie Mellon University

Exploiting Multi-Agent Interactions for Identifying the Best-Payoff Information Source

Conference Paper, Proceedings of IEEE/WIC/ACM International Conference on Intelligent Agent Technology (IAT '05), pp. 344 - 350, September, 2005

Abstract

In many different applications on the Web, distributed agents would like to discover and access high quality information sources. This is a challenging problem since an agent does not know a priori which information source would provide high quality information for particular topics. In this paper, we utilize machine learning techniques to allow a set of distributed agents to use their past experience and collaborate with others to identify information sources with the best payoff. The proposed method allows an individual agent to estimate the next payoff based on its own history of interactions with the information source and also on collaboration with other agents whose individual analysis of the next payoff the agent trusts. Q-learning is applied for stochastic updates to the payoff. Experimental results show that the proposed method provides the best results when an individual agent collaborates with a moderate number of neighbors.

Notes
IAT-2005, http://www.hds.utc.fr/IAT05

BibTeX

@conference{Seo-2005-9280,
author = {Young-Woo Seo and Katia Sycara},
title = {Exploiting Multi-Agent Interactions for Identifying the Best-Payoff Information Source},
booktitle = {Proceedings of IEEE/WIC/ACM International Conference on Intelligent Agent Technology (IAT '05)},
year = {2005},
month = {September},
pages = {344 - 350},
publisher = {IEEE Computer Society Press},
keywords = {Multi-Agent Interaction Modeling, Multi-Agent System, Machine Learning, Q-Learning, Intelligent Software Agent},
}