An Abridged Review of Transfer Learning Technology

Computer vision studies have adapted several deep learning approaches to achieve desired results. One of such approaches is transfer learning, which uses pre-trained models as a starting point for computer vision tasks. A model is trained with particular dataset to recognize certain patterns. There are lots of models namely ResNet, VGG-16, MobileNet, Inception etc. These models are trained on more than a million images from the ImageNet database. By using transfer learning technology a model can be reused in a new problem. Typically, in computer vision, neural networks try to detect edges in the first layer, shapes in the intermediate layer, and task-specific properties in the latter layers. Transfer learning uses the early and central layers, while the subsequent layers are only retrained. Transfer learning performs several beneficial functions, including reducing training time, improving neural network performance, and reducing dataset size. Transfer learning has become a promising and popular deep learning field because of its extensive application possibilities. This survey analyses the existing transfer learning research studies, which can help upcoming researchers to better understand current research level and concepts in transfer learning. A review of 20 transfer learning approaches is presented in this survey.