learn exhaust train dataset case studi detect bacteria cultur optic scatter technolog murat dundar daniel hirleman arun bhunia paul robinson bartek rajwa train dataset nonexhaust list class class repres result learn problem ill defin case sampl miss class incorrectli classifi exist class applic cost misclassifi sampl neglig signific problem acknowledg potenti undesir consequ incorrectli classifi food pathogen nonpathogen consid research direct real time detect food pathogen optic scatter technolog bacteri coloni consist progeni singl parent cell scatter light nm produc uniqu forward scatter signatur spectral signatur descript characterist bacteri coloni identifi bacteria cultur real time bottleneck remain address nonexhaust natur train librari difficult impract collect sampl bacteria coloni construct digit librari exhaust set scatter signatur studi deal real time detect sampl miss class problem learn nonexhaust train dataset propos method assum common prior set class miss paramet prior estim sampl class prior gener larg number sampl simul space miss class final bayesian maximum likelihood classifi implement sampl real simul class experi perform sampl collect bacteria subclass favor propos approach state art 