Disagreement-Based Active Learning in Online Settings

We study online active learning for classifying streaming instances within the framework of statistical theory. At each time, learner either queries label current instance or predicts based on past seen examples. The objective is to minimize number while constraining prediction errors over a horizon length $T$. develop disagreement-based algorithm general hypothesis space and under Tsybakov noise establish its complexity constraint bounded regret in terms classification errors. further matching (up poly-logarithmic factor) lower bound, demonstrating order optimality proposed algorithm. address tradeoff between show that can be modified operate at different point curve.