Professional Documents
Culture Documents
ILP = Inductive Logic Programming = machine learning logic programming = learning with logic
Introduced by Muggleton in 1992
(Machine) Learning
he proce!! by which relatively permanent change! occur in behavioral potential a! a re!ult o" e#perience$ (%nder!on) Learning i! con!tructing or modi"ying repre!entation! o" what i! being e#perienced$ (Michal!&i) % computer program i! !aid to learn "rom e#perience E with re!pect to !ome cla!! o" ta!&! T and per"ormance mea!ure P' i" it! per"ormance at ta!&! in T' a! mea!ured by P' improve! with e#perience E$ (Mitchell)
car8propertie!
*ar c11 c12 c17 c16 c21 / Length !hort long !hort long !hort / 2hape %#e! ,oo" none none pea&ed none "lat / / / / / / / / rectangle 2 rectangle 7 rectangle 2 rectangle 2 rectangle 2 / /
Mo!t ML techni(ue! can:t u!e more than 1 relation e$g$9 deci!ion tree!' neural networ&!' /
Prolog + de"inition!
<ariable!9 X, Y, Something, Somebody erm!9 arthur, 1, 1,!,"# Predicate!9 father$!, female$1 =act!9
father(%hri&topher,'i%toria). female('i%toria).
,ule!9
parent(X,Y) :- father(X,Y).
Find9 a theory T (or hypothesis) !uch that ; >+ c(e)' "or all e 3
E9 27A molecule! with &nown cla!!i"ication C9 a%ti'e and nona%ti'e w$r$t$ mutagenicity Po!!ible T9
a%ti'e(1ol) :atom(1ol,2,%,!!), atom(1ol,3,%,1*), bond(1ol,2,3,1).
%!!
%1*
Learning a! !earch
Given9
4 4 4 4 4 ;ac&ground &nowledge ; heory )e!cription Language T Po!itive! e#ample! P (cla!! B) -egative e#ample! - (cla!! +) % covering relation covers(B,T,e)
Learning a! !earch
*overing relation in ILP cover!(;' 'e) ; >+ e % theory i! a !et o" rule! 3ach rule i! !earched !eparately (e""iciency) % rule mu!t be con!i!tent (cover no negative!)' but not nece!!ary complete Separate-and-conquer !trategy
4 ,emove "rom P the e#ample! already covered
2pace e#ploration
2trategy1 ,andom wal&
4 ,edundancy' incompletene!! o" the !earch
.enerality ordering
,ule 1 i! more general than rule 2 =C ,ule 1 cover! more e#ample! than rule 2
4 I" a rule i! con!i!tent (cover! no negative!) then every !peciali!ation o" it i! con!i!tent too 4 I" a rule i! complete (cover! all po!itive!) then every generali!ation o" it i! complete too
Mean! to prune the !earch !pace 2 &ind! o" move!9 !peciali!ation and generali!ation *ommon ILP ordering9 D+!ub!umption
.enerality ordering
parent(X,Y):-
parent(X,Y):- female(X)
parent(X,Y) :- father(X,Y)
&pe%iali&ation
%on&i&tent rule
2earch bia!e!
!"ias re ers to any criterion or choosing one genera#i$ation over another other than strict consistency with the o%served training instances&' (Mitche##)
,e!trict the !earch !pace (e""iciency) .uide the !earch (given domain &nowledge) )i""erent &ind! o" bia!
4 Language bia! 4 2earch bia! 4 2trategy bia!
Language bia!
*hoice o" predicate!9
roof(-,flat) 4 roof(-) 4 flat(-) 4
2earch bia!
he move! direction in the !earch !pace op+down
4 !tart9 the empty rule (c(?) 9+ $) 4 move!9 !peciali!ation!
;ottom+up
4 !tart9 the bottom clau!e (F c(?) 9+ ;$) 4 move!9 generali!ation!
;i+directional
2trategy bia!
Geuri!tic !earch "or a be!t rule Gill+climbing9
4 5eep only one rule 4 e""icient but can mi!! global ma#imum
;eam !earch9
4 al!o &eep k rule! "or bac&+trac&ing 4 le!! greedy
;e!t+"ir!t !earch9
4 &eep all rule! 4 more co!tly but complete !earch
initialiEe a !et o" theorie! a! the !earch !tarting point! Sele%t(Rules,Examples): !elect the mo!t promi!ing candidate rule R :efine(R,Examples): return! the neighbour! o" R (u!ing !peciali!ation or generali!ation) :edu%e(Rules,Examples): di!card unpromi!ing theorie! (all but one in hill+climbing' none in be!t+ "ir!t !earch)
ILPnet2 4 www$c!$bri!$ac$u&HFILPnet2H
-etwor& o" 3#cellence in ILP in 3urope 7I univer!itie! and re!earch in!titute! 3ducational material! Publication! 3vent! (con"erence!' !ummer !chool!' /) )e!cription o" ILP !y!tem! %pplication!
ILP !y!tem!
=JIL (Kuinlan and *ameron+Lone! 1997)9 top+down hill+climbing !earch Progol (Muggleton' 199M)9 top+down be!t+"ir!t !earch with bottom clau!e .olem (Muggleton and =eng 1992)9 bottom+up hill+ climbing !earch LI-N2 (Lavrac and )Eero!&i 1996)9 propo!itionali!ation %leph (FProgol)' ilde (relational deci!ion tree!)' /
ILP application!
Li"e !cience!
4 mutagenecity' predicting to#icology 4 protein !tructureH"olding
3ngineering
4 "inite element me!h de!ign
3nvironmental !cience!
4 biodegradability o" chemical compound!
he end
% "ew boo&! on ILP/ L$ Lloyd$ Logic "or learning9 learning comprehen!ible theorie! "rom !tructured data$ 2AA7$ 2$ )Eero!&i and -$ Lavrac' editor!$ ,elational )ata Mining$ 2eptember 2AA1$ L$ )e ,aedt' editor$ %dvance! in Inductive Logic Programming$ 199O$ -$ Lavrac and 2$ )Eero!&i$ Inductive Logic Programming9 echni(ue! and %pplication!$ 1996$