Exploiting Qualitative Knowledge in the Learning of Conditional Probabilities of Bayesian Networks

Algorithms for learning the conditional proba­ bilities of Bayesian networks with hidden vari­ ables typically operate within a high-dimensional search space and yield only locally optimal so­ lutions. One way of limiting the search space and avoiding local optima is to impose quali­ tative constraints that are based on background knowledge concerning the domain. We present a method for integrating formal statements of qual­ itative constraints into two learning algorithms, APN and EM. In our experiments with synthetic data, this method yielded networks that satisfied the constraints almost perfectly. The accuracy of the learned networks was consistently superior to that of corresponding networks learned without constraints. The exploitation of qualitative con­ straints therefore appears to be a promising way to increase both the interpretability and the accu­ racy of learned Bayesian networks with known structure. If you don 't know where you're going, you might wind up someplace else. -Yogi Berra