Speaker:Nikhita Sharma

Title: Hierarchical Clustering based Structural Learning of Bayesian Networks

Abstract: Bayesian networks are being used in various domains, such as data mining, diagnosis, bioinformatics/computational biology, etc. One problem associated with Bayesian networks is to learn their structures from training data. In this paper, we introduce a new approach to structural learning of Bayesian networks, that combines hierarchical clustering and curriculum learning. Using the idea of curriculum learning, we learn the network in hierarchical stages. At each stage, we learn a subset of the random variables. Experiments show that this approach learns better or equivalent networks, based on different evaluation metrics, than existing curriculum learning approaches. Furthermore, it is way faster in learning Bayesian structures, compared to other curriculum-based learning methods.

ms_final_oral_s18_nikhita_sharma.pdf