bottom dup learn markov network structur structur markov network typic learn top search step search special featur conjoin variabl featur improv score ineffici test featur variat support data highli prone local optima propos bottom search altern inspir analog approach field rule induct blm algorithm start complet train long featur repeatedli gener featur match nearest exampl drop variabl extens empir evalu demonstr blm faster accur standard top approach outperform state art method deep transfer dorder markov logic standard induct learn requir train test instanc distribut transfer learn seek remov restrict shallow transfer test instanc domain distribut deep transfer test instanc domain predic human routin perform deep transfer learn system capabl paper propos approach base form order markov logic algorithm discov structur regular sourc domain form markov logic formula predic variabl instanti formula predic target domain approach successfulli transfer learn knowledg molecular biologi social network web domain discov pattern includ broadli properti predic symmetri transit relat predic form homophili integr approach featur invent model construct drug activ predict present machin learn approach qsar task predict bind affin molecul target protein base structur approach predict bind affin regress substructur discov relat learn make contribut state art multipl instanc mi regress repres molecul set conform model activ relat learn compon emploi score sayu method select substructur abil improv regress model applic sayu multipleinst real valu predict evalu approach task demonstr sayu outperform standard coverag measur select featur regress ii mi represent improv accuraci standard singl featur vector encod iii combin sayu mi regress accur qsar approach chang represent statist relat learn statist relat learn srl algorithm learn statist model relat data store relat databas previous introduc view learn srl view relat databas automat modifi yield accur statist model present paper present sayu vista algorithm advanc initi view learn approach wai learn view introduc relat tabl field exist tabl databas tabl field limit approxim target concept approach perform type predic invent approach avoid classic problem predic invent learn useless predic keep field tabl predic immedi improv perform statist model retain field tabl definit field tabl evalu view learn approach relat classif task effici approxim lookahead relat learner relationship precis drecal roc curv receiv oper characterist roc curv commonli present result binari decis problem machin learn deal highli skew dataset precis recal pr curv give inform pictur algorithm perform show deep connect exist roc space pr space curv domin roc space domin pr space corollari notion achiev pr curv properti convex hull roc space show effici algorithm comput curv final note differ type curv signific algorithm design pr space incorrect linearli interpol point algorithm optim area roc curv guarante optim area pr curv integr approach learn bayesian network rule view learn statist relat learn applic mammographi statist relat learn srl construct probabilist model relat databas kei capabl srl learn arc bay net sens connect entri row relat tabl tabl srl approach constrain exist databas schema databas applic user find profit defin altern view databas effect defin field tabl field tabl highli learn provid srl capabl learn view view learn statist relat learn applic mammographi statist relat learn srl construct probabilist model relat databas kei capabl srl learn arc bay net sens connect entri row relat tabl tabl srl approach constrain exist databas schema databas applic user find profit defin altern view databas effect defin field tabl field tabl highli learn provid srl capabl learn view mass modular architectur sensor system method interact modular architectur sensor system mass friendli object track foreign object detect local sdac wireless sensor network 