Multi-source Transfer Learning Based on the Power Set Framework

Abstract Transfer learning is a great technology that can leverage knowledge from label-rich domains to address problems in similar lack labeled data. Most previous works focus on single-source transfer, assuming the source domain contains sufficient data and close target domain. However, practical applications, this assumption hardly met, exist different domains. To improve adaptability of transfer models for multi-source scenarios, many existing methods utilize commonality specificity across They either map all with into common feature space or combine multiple classifiers trained pairs each form classifier. correlations bring significant impacts performance are ignored. In light this, we propose novel method based power set framework (PSF-MSTL). First, PSF-MSTL constructs enables be interrelated. Second, makes source-domain integral able provide complementary using dual-promotion strategy. Additionally, formulated as an optimization problem, iterative algorithm presented it. Finally, conduct extensive experiments show outperform advanced methods.