A method of active learning with optimal sampling strategy

We present a method of active learning with optimal sampling strategy. The iterated process for training a classifier of active learning is considered as an optimal problem which consists of the classifier optimization and the sampling optimization. Our proposed algorithm is implemented with importance weighted for the linear classifiers under the general loss function. The experiments on the problem of remote sensing show that the number of the labeled data can be reduced effectively by our algorithm. Our proposed algorithm is compared favorably to the existing methods, such like passive learning and uncertain-based active learning.