SMARTER SOLUTIONS AND APPLICATIONS USING NLP TRANSFER LEARNING – A REVIEW

Abstract

Transfer learning is a machine learning research issue that focuses on retaining knowledge obtained while resolving one problem and transferring it to another that is unrelated but still pertinent. This technique can be used with a different machine learning models, including deep learning models like reinforcement learning and artificial neural networks. The model was trained on a different set of data and issue domain from the one at hand. But there might be some connections between the two application domains. Consider the case when you have a dataset of dog image. During the detection of dogs, you can create a machine learning model. Additionally, horses or any other animal can be identified using the same model. The classic approach of machine learning cannot always be used because it is difficult to obtain labelled data in adequate quantities. As a result, Transfer Learning (TL) offers us the answer to numerous real-world issues. Actually, knowledge is transferred between domains using transfer learning. Transfer learning approaches can be partitioned into two categories: homogeneous and heterogeneous. While learning involving problem domains from the similar feature space is referred to as homogenous transfer learning. Domains so only vary in marginal distributions. Contrarily, heterogeneous transfer learning is useful when the property of the domains is different. Hence the result, it is more challenging than homogeneous transfer learning.