activ learn network data introduc activ learn algorithm classif network data set train instanc connect set link form network label link node correl goal exploit depend accur label node problem aris domain includ social biolog network analysi document classif recent interest method collect classifi node network case label exampl expens network inform show activ learn algorithm advantag network structur algorithm effect exploit link instanc interact local collect aspect classifi improv accuraci learn fewer label exampl experi real world benchmark collect classif domain show achiev extrem accur result small fraction data label learn disambigu search queri short session transfer learn minim target data map relat domain central goal transfer learn enabl learn train data domain interest limit work transfer relat domain focus case signific amount target data paper bridg gap studi transfer amount target data minim consist inform hand entiti extrem case singl entiti present sr lr algorithm find effect map predic sourc model target domain set render preexist knowledg target task demonstr sr lr effect benchmark relat domain social interact studi behavior inform increas number entiti map revis markov logic network transfer learn transfer learn address problem leverag knowledg acquir sourc domain improv accuraci speed learn relat target domain paper consid transfer learn markov logic network mln power formal learn relat domain present complet mln transfer system autonom map predic sourc mln target domain revis map structur improv accuraci result real world domain demonstr approach successfulli reduc amount time train data need learn accur model target domain learn scratch bottom dup learn markov logic network structur markov logic network mln statist relat model consist weight firstord claus gener order logic markov network current state theart algorithm learn mln structur top paradigm potenti candid structur systemat gener data evalu statist measur fit data exist algorithm outperform impress arrai benchmark greedi search suscept local maxima plateau present algorithm learn mln structur bottom approach address problem algorithm proposit markov network learn method construct templat network guid construct candid claus algorithm significantli improv accuraci learn time exist topdown approach real world domain activ reloc aid reinforc learn experi ensembl miss noisi data potenti advantag multipl classifi system increas robust nois imperfect data java teach network concept programm network sniffer 