A Concise Review of Transfer Learning

TL;DR

This paper provides a concise review of transfer learning, highlighting its ability to improve model performance across different domains by utilizing related source data, and discusses current challenges and approaches.

Contribution

It offers a comprehensive overview of traditional and current transfer learning settings, challenges, and related methodologies, summarizing the field's evolution and key issues.

Findings

Transfer learning enables models to perform well across different domains.

It addresses challenges related to domain discrepancy and data scarcity.

The survey summarizes existing approaches and future directions.

Abstract

The availability of abundant labeled data in recent years led the researchers to introduce a methodology called transfer learning, which utilizes existing data in situations where there are difficulties in collecting new annotated data. Transfer learning aims to boost the performance of a target learner by applying another related source data. In contrast to the traditional machine learning and data mining techniques, which assume that the training and testing data lie from the same feature space and distribution, transfer learning can handle situations where there is a discrepancy between domains and distributions. These characteristics give the model the potential to utilize the available related source data and extend the underlying knowledge to the target task achieving better performance. This survey paper aims to give a concise review of traditional and current transfer learning settings, existing challenges, and related approaches.