PROJECT TITLE :

Optimal Bayesian Transfer Learning - 2018

ABSTRACT:

Transfer learning has recently attracted important research attention, because it simultaneously learns from different supply domains, that have plenty of labeled information, and transfers the relevant knowledge to the target domain with restricted labeled data to improve the prediction performance. We tend to propose a Bayesian transfer learning framework, within the homogeneous transfer learning state of affairs, where the supply and target domains are connected through the joint previous density of the model parameters. The modeling of joint previous densities permits higher understanding of the “transferability” between domains. We tend to define a joint Wishart distribution for the precision matrices of the Gaussian feature-label distributions in the source and target domains to act like a bridge that transfers the helpful information of the supply domain to assist classification within the target domain by improving the target posteriors. Using several theorems in multivariate statistics, the posteriors and posterior predictive densities are derived in closed forms with hypergeometric functions of matrix argument, leading to our novel closed-kind and fast Optimal Bayesian Transfer Learning (OBTL) classifier. Experimental results on each artificial and real-world benchmark knowledge confirm the very good performance of the OBTL compared to the other state-of-the-art transfer learning and domain adaptation methods.