A survey on Deep Transfer Learning for Convolution Neural Networks
Abstract
Deep Learning approaches learn complex patterns from the training data without hand-crafted features. In this way, Deep Learning models require lots of data to train the model, which is essentially very expensive and time-consuming. Convolutional Neural Networks (CNN) is one of the Deep Learning approaches. Moreover, CNN needs labeled training data. Transfer learning has recently become very popular as a new approach for addressing this issue. Transfer learning focuses on enhancing the performance of the target task in the target domain by utilizing the knowledge acquired from the related source domain. Transfer learning reduces the data dependency to accomplish the target task. A literature survey has been presented on existing transfer learning approaches in machine learning, deep learning, and convolutional neural networks in a comprehensive way. Furthermore, the advantages and challenges of transfer learning have also explained. Finally, we briefly discuss the applications of transfer learning in multiple domains.