1: \begin{abstract}

2: We present a new active learning algorithm based on nonparametric estimators of the regression function.

3: Our investigation provides probabilistic bounds for the rates of convergence of the generalization error achievable by proposed method over a broad class of underlying distributions.  

4: We also prove minimax lower bounds which show that the obtained rates are almost tight.

5: %(up to logarithmic terms) by over a broad class of underlying distributions.  

6: \end{abstract}

7: