Domain Adaptation for Gaussian Process Classification

Traditional machining learning method aims at using the labeled data or unlabeled data to train a mathematic model then it can be used to predict the unlabeled data for Data mining problem, but it requires that the data which be trained should have same distribution with the predicting data. For the real world datasets, it is hard to get enough training datasets which has the same distribution. Thus, how to train a good mathematic model by using different distribution data is crucial problem, and the researchers using the probability view to solve transfer classification problem is relative less. In this paper, we propose a transfer classification algorithm based on the Gaussian Process model, which can be used to solve the homogeneous transfer classification problem. We use the probability theory to propose a novel classification transfer learning model based on the Gaussian Process (GP) model. We experiment on the synthetic and realworld datasets and compare to other method, the result has verified the effectiveness of our approach.