Privacy Preserved and Efficient Bayesian Network Training and Inference

Abstract

Bayesian networks (BNs) are a widely used graphical model in machine learning for representing knowledge with uncertainty, and have good interpretability. BNs can perform casual inference based on partial evidence, and are useful for medical diagnosis. The learning and inference of BNs is expensive, especially for large networks, and meanwhile, the privacy of training data needs to be preserved for critical applications in domains like medical and finance. This project will research on designing efficient and privacy preserved algorithms for BNs by exploiting GPUs and CPUs.