Transfer Learning: A Paradigm for Machine Assisted Knowledge Transfer

Abstract

This paper surveys transfer learning as a sustainable knowledge transfer mechanism. Conventional machine learning algorithms require a huge amount of labeled data for supervised learning. In absence of such data, the models suffer from performance degradation. Transfer learning enables the prior knowledge gained in doing a particular task to be reused or transferred to another new task of similar nature. This can speed up and improve the learning curve of the tasks in the new domain. The paper gives an overview of the transfer learning process and highlights how this innovative artificial intelligence (AI) technique can help achieve the goals of sustainable development. The literature survey highlights widely used mechanisms of Transfer Learning like homogeneous, heterogeneous, as well as instance-based, parameter-based, and relational-based implementation of transfer learning. It discusses how these mechanisms are utilized to create efficient AI-based applications which aid sustainability in the long run.