Learning with Knowledge from Multiple Experts

Abstract

The use of domain knowledge in a learner can greatly improve the models it produces. However, high-quality expert knowledge is very difficult to obtain. Traditionally, researchers have assumed that knowledge comes from a single self-consistent source. A little-explored but often more feasible alternative is to use multiple weaker sources. In this paper we take a step in this direction by developing a method for learning the structure of a Bayesian network from multiple experts. Data is then used to refine the structure and estimate parameters. A simple analysis shows that even relatively few noisy experts can produce high-quality knowledge when combined. Experiments with real and simulated experts in a variety of domains show the benefits of this approach. ICML Proceedings of the Twentieth International Conference on Machine Learning