An Iterative Transfer Learning based Classification framework
Jihai Yang, Shijun Li, Wenning Xu · 2018
Despite of the success of transfer learning, two problems still exist that the samples used are not so informative and how to use only limited samples for labeling and for training classifiers. The active learning method may provide one solution, which involves searching for the most informative unmarked samples by query function, submitting them to the expert function for marking, then using the samples to train the classification model in order to improve the accuracy of the model and use the newly acquired knowledge to inquire into the next round, with the aim of getting the highest accuracy of classification using minimal training samples. We aim to develop a method that combines active learning strategy with transfer learning. Experimental results prove that the usage of combining active learning and transfer learning could further improve the generalization ability of classification models.