detect uniqu tempor segment inform theoret meta cluster shin ando einoshin suzuki central challeng tempor data analysi obtain knowledg underli dynam paper address observ noisi stochast process attempt detect tempor segment relat inconsist irregular dynam convent anomali detect approach detect anomali base distanc pattern provid limit intuit gener process anomali model base approach difficulti identifi small cluster set anomali propos inform theoret meta cluster itmc formal model base cluster principl theori lossi data compress itmc identifi uniqu cluster distribut diverg significantli entir dataset itmc emploi regular term deriv prefer high compress rate critic precis detect empir evalu appli itmc tempor anomali detect task dataset gener process involv heterogen inconsist dynam comparison baselin method show propos algorithm detect segment irregular state significantli high precis recal 