Optimal Combination of Information from Multiple Sources.
Abstract
A computer decision aiding system for debiasing and combining information from multiple sources (e.g. human experts, sensors) is proposed. The algorithm is based on six assumptions that apply when the sources are relately knowledgeable with respect to the operator on the variable of interest, and the operator is willing to base his evaluation of their performance on a previously selected (finite) sequence of so called calibration variables. It is also assumed that the operator is interested in maximizing gains, that is, he wishes to act in an optimal or Bayesian manner. An experiment with two human sources of information was conducted to evaluate the performance of the aiding system under a variety of loss functions. On a family of bilinear loss functions, the output of the aid was found to perform better than a naive scheme like simply believing the information the two sources gave. The combination rule was also found to perform better than the output to any individual source. Keywords: Man computer interface. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 31, 1986
- Accession Number
- ADA174726
Entities
People
- Max B. Mendel
- Thomas B. Sheridan
Organizations
- Massachusetts Institute of Technology