onlin discoveri mainten time seri motif abdullah mueen eamonn keogh detect repeat subsequ time seri motif problem shown great util higher level data mine algorithm includ classif cluster segment forecast rule discoveri recent year signific research effort spent effici discov motif static offlin databas domain inher stream natur time seri demand onlin discoveri mainten time seri motif paper develop onlin motif discoveri algorithm monitor maintain motif real time recent histori stream algorithm worst case updat time linear window size extend maintain complex pattern structur contrast current offlin algorithm signific updat time requir costli pre process step onlin algorithm simpli afford core idea extens algorithm deal arbitrari data rate discov multidimension motif demonstr util algorithm varieti case studi domain robot acoust monitor onlin compress 