OPTIMAL ADAPTIVE SEARCH.
Abstract
Optimal search strategies are developed for locating an object hidden in one of several locations. The five parameters considered in the search process are location probabilities, detection probabilities, testing costs, terminal penalty, and terminal reward for locating the object. A set of optimal adaptive-search decision-making policies can be developed that yield the minimum expected loss (a combination of the expected search cost and expected penalty and reward) when the parameters are known, and the minimum expected mean loss when the parameters are not known. In the latter case, the unknown parameters can be treated as random variables and our prior state of knowledge of these parameters can be expressed as probability distributions. Convenient prior distributions applied to these unknown random variables are given. A Bayesian learning approach is used to update the parameters through learning observations (results of past searches). With such an adaptive-search process, the decision-making policy will be improved from these learning observations and eventually converge to the true optimal search policy. The expected value of learning observations on the location probabilities provides the value of past data for improving the search process. The expected value of clairvoyance about location probability, terminal penalty, and terminal reward is the upper-bound value of these experimental learning programs. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1966
- Accession Number
- AD0804872
Entities
People
- Wesley W. Chu
Organizations
- Stanford University