Curriculum Learning of Bayesian Network Structures

Paper author list:

Yanpeng Zhao, Yetian Chen, Kewei Tu, Jin Tian

Published time：

2015

Paper title：

Curriculum Learning of Bayesian Network Structures

Paper abstract：

Bayesian networks (BNs) are directed graphical models that havebeen widely used in various tasks for probabilistic reasoning and causalmodeling. One major challenge in these tasks is to learn the BN structures fromdata. In this paper, we propose a novel heuristic algorithm for BN structurelearning that takes advantage of the idea of /emph{curriculum learning}. Ouralgorithm learns the BN structure by stages. At each stage a subnet is learnedover a selected subset of the random variables conditioned on fixed values ofthe rest of the variables. The selected subset grows with stages and eventuallyincludes all the variables. We prove theoretical advantages of our algorithmand also empirically show that it outperformed the state-of-the-art heuristicapproach in learning BN structures.

Conference or journal’s name and website：

the 7th Asian Conference on Machine Learning (ACML 2015)，http://acml-conf.org/2015/

A short description of the conference or the journal:

The Asian Conference on Machine Learning (ACML) is aninternational conference in the area of machine learning. It aims at providinga leading international forum for researchers in Machine Learning and relatedfields to share their new ideas and achievements. The 7th Asian Conference onMachine Learning (ACML2015) was held in Hong Kong on November 20-22, 2015.