In this paper, we propose a new research problem on active learning from data streams, where data volumes grow continuously, and labeling all data is considered expensive and impractical. The objective is to label a small ...