This document summarizes research on transfer defect learning to improve cross-project defect prediction. It presents Transfer Component Analysis (TCA) as a state-of-the-art transfer learning technique that maps data from source and target projects into a shared feature space to make their distributions more similar. It then proposes TCA+ which augments TCA with data normalization and decision rules to select the optimal normalization method based on characteristics of the source and target datasets. Experimental results on two cross-project defect prediction datasets show that TCA+ significantly outperforms traditional cross-project prediction and basic TCA.