mine distanc base outlier larg databas metric space yufei tao xiaokui xiao shuigeng zhou set object object outlier exist object distanc valu distanc metric provid user run time object return outlier smallest cost paper consid gener version problem inform outlier comput object mutual distanc prove upper bound memori consumpt permit discoveri outlier scan dataset time upper bound turn extrem low practic actual memori capac realist dbm typic larger develop algorithm integr theoret find carefulli design heurist leverag addit memori improv effici techniqu report outlier scan dataset case significantli outperform exist solut factor order magnitud 