Convex Batch Mode Active Sampling via α-Relative Pearson Divergence