Abstract
Structure learning algorithms usually focus on the compactness of the learned model. However, for general compact models, both exact and approximate inference are still NP-hard. Therefore, the focus only on compactness leads to learning models that require approximate inference techniques, thus reducing their prediction quality. In this paper, we propose a method for learning an attractive class of models: bounded-treewidth junction trees, which permit both compact representation of probability distributions and efficient exact inference. Using Bethe approximation of the likelihood, we transform the problem of finding a good junction tree separator into a minimum cut problem on a weighted graph. Using the graph cut intuition, we present an efficient algorithm with theoretical guarantees for finding good separators, which we recursively apply to obtain a thin junction tree. Our extensive empirical evaluation demonstrates the benefit of applying exact inference using our models to answer queries. We also extend our technique to learning low tree-width conditional random fields, and demonstrate significant improvements over state of the art block-L1 regularization techniques.
Original language | English |
---|---|
Pages (from-to) | 113-120 |
Number of pages | 8 |
Journal | Proceedings of Machine Learning Research |
Volume | 5 |
State | Published - 2009 |
Externally published | Yes |
Event | 12th International Conference on Artificial Intelligence and Statistics, AISTATS 2009 - Clearwater, FL, United States Duration: 16 Apr 2009 → 18 Apr 2009 |