Feature Spaces-based Transfer Learning

Abstract

Transfer learning provides an approach to solve target tasks more quickly and effectively by using previously-acquired knowledge learned from source tasks. Most of transfer learning approaches extract knowledge of source domain in the given feature space. The issue is that single perspective can't mine the relationship of source domain and target domain fully. To deal with this issue, this paper develops a method using Stacked Denoising Autoencoder (SDA) to extract new feature spaces for source domain and target domain, and define two fuzzy sets to analyse the variation of prediction accuracy of target task in new feature spaces.