Transfer learning: applications in image and natural language data

Abstract

Transfer learning appears to be one of the most influential techniques used in machine learning today with applications in nearly all state of the art models. From natural language machine learning to computer vision and tabular data, transfer learning has reshaped the way machine learning algorithms and models are developed and applied. In this thesis the way transfer learning works is examined and results of its applications on different domains (agricultural-image data, medical-natural language data) are presented. We conclude that this technique could change the landscape of machine learning and artificial intelligence as a whole even more within the next few years, something backed up by its performance and flexibility on a variety of tasks.