We propose an informationtheoretic ap proach for predictive modeling with Bayesian networks. Our approach is based on the min imax optimal Normalized Maximum Likeli hood (NML) distribution, motivated by the MDL principle. In particular, we present a parameter learning method which, together with a previously introduced NMLbased model selection criterion, provides a way to construct highly predictive Bayesian network models from data. The method is parameter free and robust, unlike the currently pop ular Bayesian marginal likelihood approach which has been shown to be sensitive to the choice of prior hyperparameters. Empirical tests show that the proposed method com pares favorably with the Bayesian approach
【 预 览 】
附件列表
Files
Size
Format
View
Locally Minimax Optimal Predictive Modeling with Bayesian Networks