IMPROVE A LEARNER: A SURVEY OF TRANSFER LEARNING

Abstract

Transfer learning is a subclass of machine learning, which uses training data (source), drawn from a diverse domain than that of the testing data (target). The real world is messy and contains an infinite range of novel eventualities. Transfer learning across different feature spaces is usually a tougher problem than Transfer Learning within the common feature space. This survey paper defines transfer learning to feature representation that maps the target domain to the source domains exploiting a set of meticulously manufactured features and applications to transfer learning. This paper systematically examines a feature based transfer techniques and reviews some current analysis on the topic of negative transfer