Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
Time and sample efficient discovery of Markov blankets and direct causal relations
Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining
Block Learning Bayesian Network Structure from Data
HIS '04 Proceedings of the Fourth International Conference on Hybrid Intelligent Systems
Speculative Markov Blanket Discovery for Optimal Feature Selection
ICDM '05 Proceedings of the Fifth IEEE International Conference on Data Mining
Elements of Information Theory (Wiley Series in Telecommunications and Signal Processing)
Elements of Information Theory (Wiley Series in Telecommunications and Signal Processing)
Towards scalable and data efficient learning of Markov boundaries
International Journal of Approximate Reasoning
Learning Local Components to Understand Large Bayesian Networks
ICDM '09 Proceedings of the 2009 Ninth IEEE International Conference on Data Mining
A decomposition algorithm for learning Bayesian network structures from data
PAKDD'08 Proceedings of the 12th Pacific-Asia conference on Advances in knowledge discovery and data mining
Fast Markov blanket discovery algorithm via local learning within single pass
Canadian AI'08 Proceedings of the Canadian Society for computational studies of intelligence, 21st conference on Advances in artificial intelligence
| Hi-index | 0.00 |
Markov blanket discovery plays an important role in both Bayesian network induction and feature selection for classification tasks. In this paper, we propose the Dynamic Ordering-based Search algorithm (DOS) for learning a Markov blanket of a domain variable from statistical conditional independence tests on data. The new algorithm orders conditional independence tests and updates the ordering immediately after a test is completed. Meanwhile, the algorithm exploits the known independence to avoid unnecessary tests by reducing the set of candidate variables. This results in both efficiency and reliability advantages over the existing algorithms. We theoretically analyze the algorithm on its correctness and empirically compare it with the state-of-the-art algorithm. Experiments show that the new algorithm achieves computational savings of around 40% on multiple benchmarks while securing similar or even better accuracy.