Schema Independent Rela/onal Learning Jose Picado, Arash Termehchy, Alan Fern, Parisa Ataei Informa/on and Data Management and Analy/cs (IDEA) Lab
Design a drug to treat HIV What is the structure of compounds that have an#-HIV ac/vity? A compound has an#-HIV ac/vity if it has the following substructure: Oracle N O N 2
Rela/onal learning Leverages the structure of the rela/onal • database Learns a Datalog defini/on • compound atom compId atomId atomId element c1 a1 a1 N Training data: c2 a10 a2 O an#-HIV no-an#-HIV bond compId compId atomId1 atomId2 type c1 c2 a1 a2 single c3 c4 a2 a3 single an/-HIV(x) :- compound(x,u), atom(u,N), compound(x,v), atom(v,O), Rela/onal learning compound(x,w), atom(w,N), algorithm bond(u,v,single), bond(v,w,single). 3
Rela/onal learning has many applica/ons in data analy/cs & management • Model en//es and rela/onships between en//es Marke#ng Drug design How will new customers What is the structure of respond to an offer? compounds to fight a disease? Concept Concept interestedInOffer(customer) ac/ve(compound) • Various applica/ons in data management • E.g., informa/on extrac/on, usable query interfaces, data integra/on/ exchange. 4
Benefits of rela/onal learning ü Leverage the structure of compound atom data and learn over complex compId atomId atomId element schemas with mul/ple tables c1 a1 a1 N c2 a10 a2 O ü Automa/c feature extrac/on and selec/on bond atomId1 atomId2 type ü Results are interpretable a1 a2 single (Datalog) a2 a3 single an/-HIV(x) :- compound(x,u), atom(u,N), compound(x,v), atom(v,O), FOIL, Progol, … compound(x,w), atom(w,N), Castor (new algorithm) bond(u,v,single), bond(v,w,single). Exis/ng algorithms 5
Schema 1 Which authors are collaborators ? paperAuthor author authorAffilia#on paperId authorId id name id affilia/on collaborators p1 mad mad Madden mad MIT person1 person2 p1 bai sto Stonebraker sto MIT Madden Bailis p2 soc soc Socher soc Stanford Socher Manning p2 man man Manning man Stanford Madden Stonebraker p3 mad bai Bailis bai Stanford non-collaborators paper paperYear paperConf person1 person2 id /tle id year id conf Madden Socher p1 MacroBase: Priori… p1 2017 p1 SIGMOD Manning Bailis p2 GloVe: Global Vect… p2 2014 p2 EMNLP ? FOIL learning algorithm 6
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- author authorAffilia#on id name id affilia/on paperAuthor paperId authorId paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- true. 7
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on paperAuthor paperId authorId paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- true. 8
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on paperAuthor paperId authorId paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x). 9
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x). 10
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y). 11
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId f=0 f=2 f=-1 paperAuthor(w,z) paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y). 12
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId f=0 f=2 f=-1 paperAuthor(w,z) paperConf paper paperYear id conf id /tle id year Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y), paperAuthor(w,z). 13
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId f=0 f=2 f=-1 paperAuthor(w,z) paperConf paper paperYear id conf id /tle id year f=1 f=3 f=1 paperAuthor(w,v) Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y), paperAuthor(w,z). 14
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId f=0 f=2 f=-1 paperAuthor(w,z) paperConf paper paperYear id conf id /tle id year f=1 f=3 f=1 paperAuthor(w,v) Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y), paperAuthor(w,z), paperAuthor(w,v). 15
FOIL: rela/onal learning algorithm Schema 1 collaborators(x,y) :- f=0 f=0 f=-1 author authorAffilia#on author(z,x) author(z,y) id name id affilia/on f=0 f=1 f=0 paperAuthor author(v,y) paperId authorId f=0 f=2 f=-1 paperAuthor(w,z) paperConf paper paperYear id conf id /tle id year f=1 f=3 f=1 paperAuthor(w,v) Scoring func/on f: P - N f=2 f=1 f=1 P: posi/ve examples covered No improvement N: nega/ve examples covered collaborators(x,y) :- author(z,x), author(v,y), paperAuthor(w,z), paperAuthor(w,v). 16
Schema 1 Which authors are collaborators ? paperAuthor author authorAffilia#on paperId authorId id name id affilia/on collaborators p1 mad mad Madden mad MIT person1 person2 p1 bai sto Stonebraker sto MIT Madden Bailis p2 soc soc Socher soc Stanford Socher Manning p2 man man Manning man Stanford Madden Stonebraker p3 mad bai Bailis bai Stanford non-collaborators paper paperYear paperConf person1 person2 id /tle id year id conf Madden Socher p1 MacroBase: Priori… p1 2017 p1 SIGMOD Manning Bailis p2 GloVe: Global Vect… p2 2014 p2 EMNLP f=3 collaborators(x,y) :- author(z,x), author(v,y), FOIL learning paperAuthor(w,z), paperAuthor(w,v). algorithm Two people are collaborators if they are co-authors. 17
People represent the same data using different schemas author authorAffilia#on author id name id affilia/on id name affilia/on mad Madden mad MIT mad Madden MIT sto Stonebraker sto MIT sto Stonebraker MIT soc Socher soc Stanford soc Socher Stanford man Manning man Stanford man Manning Stanford bai Bailis bai Stanford bai Bailis Stanford paper paperYear paper id /tle id year id /tle year conference p1 MacroBase: Priori… p1 2017 p1 MacroBase: Priori… 2017 SIGMOD p2 GloVe: Global Vect… p2 2014 p2 GloVe: Global Vect… 2014 EMNLP paperConf Composi/on id conf Denormaliza/on p1 SIGMOD beher performance p2 EMNLP 18 DBA
Schema 2 Which authors are collaborators ? paperAuthor author paperId authorId id name affilia/on collaborators p1 mad mad Madden MIT person1 person2 p1 bai sto Stonebraker MIT Madden Bailis p2 soc soc Socher Stanford Socher Manning p2 man man Manning Stanford Madden Stonebraker p3 mad bai Bailis Stanford non-collaborators paper person1 person2 id /tle year conference Madden Socher p1 MacroBase: Priori… 2017 SIGMOD Manning Bailis p2 GloVe: Global Vect… 2014 EMNLP ? FOIL learning algorithm 19
FOIL: rela/onal learning algorithm Schema 2 collaborators(x,y) :- author id name affilia/on paperAuthor paperId authorId paper id /tle year conference Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- true. 20
FOIL: rela/onal learning algorithm Schema 2 collaborators(x,y) :- author f=0 f=0 f=-1 id name affilia/on author(z,x,v) author(z,y,v) paperAuthor paperId authorId paper id /tle year conference Scoring func/on f: P - N P: posi/ve examples covered N: nega/ve examples covered collaborators(x,y) :- true. 21
Recommend
More recommend
