Risk-optimal active learning with informative abstentionwith Geoffrey J. McLachlan1 versionW_xtujafzwv1 · current