Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Exploitation vs. exploration: choosing a supplier in an environment of incomplete information
Azoulay-Schwartz R., Kraus S., Wilkenfeld J.  Decision Support Systems 38 (1): 1-18, 2004. Type: Article
Date Reviewed: Jan 28 2005

What is the best strategy for a purchaser choosing suppliers, when the quality of each is described by a distribution function? Should a purchaser deal with a new supplier, without a history, who may be better than existing suppliers? The authors’ approach is based on Gittins indices [1], but is generalized to stochastic buying intervals, different supplier prices, and different size purchases.

A helpful example demonstrates the approach. When a new supplier appears, Gittins indices cannot be calculated, because the history of the supplier is not known. Several strategies are suggested and simulated. The trade-off is between using present information (exploitation), and learning about new suppliers (exploration). One strategy is to calculate the index of the new supplier (using a short history implying a high index) from the average performance of the other suppliers. Another strategy is to select new suppliers randomly. A third strategy is to always choose a new supplier if one appears, and a final strategy is to never choose a new supplier.

The results of the simulations are reasonable. With one known supplier, and one new supplier, it is best to explore and try the new supplier. With four or five known suppliers, it is best to ignore new suppliers. With two known suppliers, it is best to assign the new supplier the average performance. The paper is wordy, but effectively demonstrates how to apply the approach to particular problems.

Reviewer:  B. Hazeltine Review #: CR130726 (0508-0948)
1) Gittins, J.C. Multi-armed bandit allocation indices. Wiley, New York, NY, 1989.
Bookmark and Share
  Reviewer Selected
Featured Reviewer
Intelligent Agents (I.2.11 ... )
Knowledge Acquisition (I.2.6 ... )
Electronic Commerce (K.4.4 )
Learning (I.2.6 )
Would you recommend this review?
Other reviews under "Intelligent Agents": Date
Bi-level thresholding: analyzing the effect of repeated errors in gesture input
Katsuragawa K., Kamal A., Liu Q., Negulescu M., Lank E.  ACM Transactions on Interactive Intelligent Systems 9(2-3): 1-30, 2019. Type: Article
Mar 24 2021
Intelligent systems for geosciences: an essential research agenda
Gil Y., Pierce S., Babaie H., Banerjee A., Borne K., Bust G., Cheatham M., Ebert-Uphoff I., Gomes C., Hill M., Horel J., Hsu L., Kinter J., Knoblock C., Krum D., Kumar V., Lermusiaux P., Liu Y., North C., Pankratius V., Peters S., Plale B., Pope A., Ravela S., Restrepo J., Ridley A., Samet H., Shekhar S.  Communications of the ACM 62(1): 76-84, 2019. Type: Article
Mar 28 2019
A scalable preference model for autonomous decision-making
Peters M., Saar-Tsechansky M., Ketter W., Williamson S., Groot P., Heskes T.  Machine Learning 107(6): 1039-1068, 2018. Type: Article
Oct 12 2018

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright © 2000-2022 ThinkLoud, Inc.
Terms of Use
| Privacy Policy