Research Frontiers in Transfer Learning -- a systematic and bibliometric review

摘要: Humans can learn from very few samples, demonstrating an outstanding generalization ability that learning algorithms are still far reaching. Currently, the most successful models demand enormous amounts of well-labeled data, which expensive and difficult to obtain, becoming one biggest obstacles use machine in practice. This scenario shows massive potential for Transfer Learning, aims harness previously acquired knowledge new tasks more effectively efficiently. In this systematic review, we apply a quantitative method select main contributions field make bibliographic coupling metrics identify research frontiers. We further analyze linguistic variation between classics frontier map promising directions.