Research Of Transfer Learning Algorithm Based On Instance

When the distribution is changed in the traditional machine learning, most of machine learningmethod is not have adaptive ability, they need to relearn, it requires the user to collect more trainingdata. In real applications, re-collect training data and re-training the learning machine are costly. Itis necessary to reduce this part of re-learning, research of transfer learning has become inevitable.Transfer learning can be applied to related data. We transfer useful knowledge of related fields tothe target field to solve the learning task in the target field. While in the transfer learning how toselect the related data to help target data learning is the key to transfer learning. This paper proposesa transfer learning algorithm based on clustering ensemble to find instance which can help targetdata study from the secondary data to help target areas of learning. And for the defect of transferlearning algorithm based on clustering ensemble is that target data is less and easy to overfitting,presented on a clustering ensemble transfer learning algorithm based on semi-supervised to improvethe defects. Finally, experiment of two algorithms on20newsgroups data set, proof of these twoalgorithms can effectively improve the learning efficiency of the target data.