HWP Document File V3.00 7R":bA ͡G)ccccbЉA ͕2002e 6 20 a 8:08:09W}tTezM@>D4."ZGY5|/U(V<o% ft3 $MP$3KqǮ;֤V?Vxiy罁ᐙ}}~w8)>>1ևY|ҟW-GmeŻ˷n 0ߩ\j|r=ꬳ}=Ӊ@V \?ݕ8p $57 ;)@}gӀfgo,,=,\>~xYkpv2+e]~y #<rz͆J/ޏf]j.gX VJbVD{r"~ZsggRNͩɾN,v0̃Za7grv[{̹{Nmƹڝ9{E1vOƌ}f`&&6s]x)@~\c,_^ju?Æe da~zTSidgw큷#f"3t|-,义 O'Mi4P<~%f!I3zL1r*m|aIYLwmʊ fU׫I6 w[kI36Oj/bӈQ~FaE:-;Uĝa'~6=; [dvjM4E]Y΃鯀e(#ځa ]F{5->gې'ruːR-N[B`EW[(.Z]wwW\gv4Ç[!e wh $m[3(]xjx*bO+ )F< 1fW٘C[:xnz:jCֲIlis7U[kQO? ,py5 8` [E2`.`o n^~m;Ez;ډY r (-Qu63=> rJkf~%k*M`q:q-* [Y=!^uN:Hs ՇA%U`TfVL; !PJSEi1vSF&tjq P gB e+-DAc- -ŞKl^y SWqފP%> I 2IsxEk;7ɣ~4\"vAJIJ PmuO2aRV K("MYS*֕!lJBeh$zuy1i~囷jkҳGݥ^$ٕ؜b_.,ݷ_r}vF6'9hEwkY k?vdһ%龳}d! >sM^dzuUrW2Zj?yNǬL+ګ3&/rM:QSKg:Oj֯w7=3Йp6u=i޷*)>urɉs_R*>uQdw .⡀~nn.LLk^DɠX؏[xiM^37@`|*GЭ Dr`X'@t]+mۤWavdo2cp#Y7BP*HkVKs.vYF|J/?G\>̯+'ȱ3qg Uha"%7Fy௅Be!ϭ^txNiܝ*=BDQP0үRiќ9E6L'}]_nJݣҿi}*}Q4 4 L_3,A5ġ ǏΗSեE9Xj6#v]$n5Gز# :hǃa %n ֱ i$lOmT[s/;ۜV҉Ye%uMG®1,Fo03ٌ*w=;Vsm[8dIk5 M:17&vh7~F+" AHf٧MV\մw|F' a`Ԭ"uL[c FI%1y~)gHKd=WPXUvVu, ֲ#X{dCS6V'~;ݝJPZ3S0OfFVe#Q$Bޞ;@'W韱4< eJxhJ8<;_z\TVR?Y4T:]J!nrV6je&sR0eoɕ9lHw]]LJ:z,xEXQa€+-=d 䖸"F]ÁXaes}\-G>Hwi!҃‘_ⓁtO=^\v$c6-E'}D bO3th KrH$EQDY:NJZYxV@f{w4t+^T9_ ܟ0Zi.e՗l$N'{(y3GTjñXiRumm;ZhibX?!ԪiW0/0l~ߤg<OsF;ojh-w $WFTf/6< ^\|V<>TmNҕd JM{?F?_j~ xḑt*-08(C #*m5XR9)=Er Qun[ug&{ATOY"_w盟G2@[[#3d?^:^/\n ɏ0\[_O9 'vw}b.ĉj`Yn9vLc%ZmD=%wRtDC&u;݈eIR J&;)-zcS2WfZkjzH᥹<]ps]E2myY+ -m|9Aɔ$c؎,^u2PTO*m[\YOIKޱw5TZБM+2dozk9L4ċU7\g{Iף? dPơTyNwvJT#-4c޻Uc3?E"cW4xB;Y6s"u+T3!/IK ֯zD'.&SoXk\^o d 4v 5mQ0<\tbљcw_ ez~oFP&x׉P)̧*mGR,HfTcև *CSV3>H}&80m};֑{nѸ15&;r?C R"r#Q m*%s ׸=U֫U̲^nㄐT4&(m[q1Y fids1rH~*¸GEgM^A+fZ@-6` V_lc+E~cMٴÕ*&dA?37>91>@C15c;A`ŋ7Z+ZV|ܯ/Q~\GiR0#^R*a`."w?pJo (hN?KT;(#&ҌxyJ+c(}|J/S'ޠeL5?K\RBc~J4T:q9s>˫{nχاjM4$5iv9JA=CǬFsKXɧ5uNL]^%\tm'Rit9S"P 좨}אmhX-Vkb$^R&<#z&vJZuP|X^K}Ltc~m'8"Io7m:FY-شlzfvrt^+̟vm蔓yeKzYu:Fo ꜸCжoR+%{k $ ۧ jxI=fg%†%ɞGO2*/#gC8OeRtHå+;N\D]_ZG]6UrLdf?6MN9V*V:mW}-U΅y)=&{V֬.0!]XG"_ x"0`]{SL9Eݤ/=$WUy@vx( tcC?H,$®ϭU7i&c2 XS_S[x_SO0Rx* cvJ?gv'ԁRvJ|J»mJbe)K$R$?PAw*^—Zjo.ު@vCg.Gik!Y,JFO0Qo)}lFJG8ӑM~Ow?!s+_5Q]6e6?mpqH$q%?XB $eF!D=57ifU:z&D)=߹wvZ龕lb?iw߹s=s; 1S/`wBV.)3zShZY/ _6YGl{UK~&x_x)/E\.ͅڸU>I# b@C26M3hh.BB@CKZ h 5)@CO-m2L!v@[@C{Қ2ħb %q ,p[iTy~ygϐo h/WA= 4 BC@#;F9 psw -3B YS5Ak\gQp\ M+یl|)Xiy"rW+>H!*oQ#J s-_T9Y{!Yê_M3W#/?rQ/F/lY)#WuфKڸX@2iɳBR:;^uX2p63cK++ZA^DD.'0&O VxZ,ulG;FR}p_,sA+AAI=&Euuhyg@@;u+m9RfY:Ӎ[|5X@{vcю,^QW,p3weO%8/7[ g$O 1@Lb|"@Ty@LS}ע†xm 4Xhc1 Dm3  ʀvUѽĶʫ߃Q?< Ӣ .NhlpBJYn'O}aeNәÏ6鋒 :H!g1J;4Գ_BLf>&]&_8P!И$FIʥnI@~pdqm?6jղRR;q41~[ϴvm$(}1gT5AFs"BţGE9ȷ=T}ju׫9sb;Y%, o:k6 ßNl8\ԫғaVM C@t&fAW'H0Ql`TAPw8q+$l{@q+a愱_69X5 Xf ]ov7]mjש[N!g֓/G.rFĺʠ_w}jzpVeGL̀i`iOp_lRͻ[:;lmk;ނ]Yx]$Bb~']Lάc?F:с~`Iz:>Kl6ڒRe0l`Y'rm+8^uwV芏t5|+oC2B72|zyf{A%ߪꬸh4jZ)yrU2TtŃY:yqU[e׃kF\\︺!5'< '.pR0b kxFȼFհ40Nc<&pzOc*`|2|[}ǜ4db$2Vɑ>eB_a\ߊq}+HoDz:hbtE4OH6e<3yNbb{)@_2={\ *TߝdVCbɳɖ +1M"P&Jj@5IR)T#k:LcA8WR4ԧwJ1!8PO̒Ѡ^OR\c@z@Ruֹ%@s%#2P/ϓT|w>pqE^PRz;GUxpISg2q*tz8:2Xo d&X= ۧB^GdޙV2=nuWDj+3>,@Ġ){]|5Qt)7?lҰ,ɂ')7?Ib]UҜ2[-@ ^[Eg҆R|N.PEj^]jAjʲ8ST)y֑AܭѠNzg/7TeѠN$+d`>dHǵ$yȨ[.:fuB/ER)[vltUq~Oڡמc|9fi6 FP#3R8+̈iyO;ީ`eOx:my6*mFUëV^kTia]oªNj 4CɐU+$}r lky^w/w.i:ʲ[w'O_Oapn;qZٍ{"S:XbPN|%85$^XOp_@@uaiM J;H,::{-R)"5#Uz f+6B y֧z#n7wy#4YMϗc6y{`8R#FQ#hNu;_v׻Ս%6qY{ "mIGR=>/Wt+ 8GS ;vzz;2tH4|}7B @U6R̙;yoَugx2fdٞWW[#_{ ٥O犏wp)c^ XzAY|Ld;^vwwP拌 B'Ϻ9qb'a.~۹򱷗xO<1r4:)+L]'2(| Q"d6KT4(D%z!Qg!닉,%8ѶDkW=hSD'zi^3jM*3ۉ(oOZR]%):+F*K0'7`N5/gVR5?])(\=:3aLZ-K")0S"1:)U%u]p#%i*$MG6bj7V/Th.>*7hUYLMK@5/XaP 1uVjs[Skh#A9SM IY+FWJJ {8 5GBWl Rs(mQ*FSlP)Ԛ8ZdNx4 Eq&QFkPݮ, F(V8` #&$H+WJvictwpYgܿ&@@ƺ,F^t\ԒѷmO uQc|`tKlc[FJV'^bRO5USEkǍZJ3=B.vfE[N#X}S+:;"SlwwRϐ}~5]yFd_Ϊ]іGp>nNFj7v4 o{ѿN҄ʖ cO巎hg2Z"G}}o8}ډNFtMU>;*ðB;3gfӟ}&.Ȯil,ζ;mc~5ڨSe' 1Vu*Gwo^GNn߻Dמ<=4}חO䁒!͵ksbQ06*{mp*k'?T.",ȶWI|@.ͅҵU>)$ /aDK"d6KT ʟf}
[R5?[8މ{yz(jSz/'.MNr9e-_~o9f~xdu}sgkߜ/UOι5˂k ,gq9p:X;*{)l-ݜDfONyX0Ι)Sf3<1Hrש ҷKurޟV:y4]5>)# :hb]OpKڮvʙu"y댽9Qg+PF:eAD PJt:!.o^n9`DxΈ}FhlNLN' a5Jq%$RӜCEٜH1H,/ӂ݋cU<*0b;V1ӂ'3"kY,|nF ӜO$.^-bސјNκrݭwf1z\ L'>F%]Ė튌"kFd %պ[3={:S\Jf6nUBI l3BB`1;Ѽn0os 21*鶕`!*l6D8 .㍼ə\@BB #VF/N.|vI/{ /vų~R?u8IIE/2t<PbA a2NdIKD+>1Ѻޠ~0C+Ovn22l_CeR@ đa2q2< 𙇐9+5^3"nNt͹-{:{!g;bMQ,.rqO\.ͅzY>ȧց>A6LnE #D1}4S.a=|z /Nr$ڵky~\ c TN-&_ QKTD:dÿ?ç/N9 "toaשBt ,Ewoϋx)AI+ .( t^(Pe\Gs5H:9-4h˧vNq#7/ ׈P(? |^_ȗGNQ$*W &(kV!͔˔7Eq")HS! '&/Hp̩YncbEFdO3Ѭ=cٯ7\ESeu.UlKKԐ4f $ύKZ%Q,*V |J#titaT.-.L-֥&RcN:]zolcN:]ܘNLkI^ ce 0\0Y2n- ^Dxg}<p!Qg!닉,%8ѶDkW?Un!C'C@P Kth )ڑ80Pb2¸6@5Uű*wjƵe˭uw}N}ܛ<Ƚєޮ9;@dhȻ4<6DvA5"KֈX'Qd1;>i)4~Rd$qn@@HuVNA" r4Dj 9ñܰX(zIO$%w3q[Vgmr<u$"sP˨00Kn>|u`"orډ^ZgݵㅽKW7߫z4h8,TƸ 'Q)FTtR ', l+R\t dBI).ƩBSRpkbʵГHp/Tru+:X)mF'}gmc> Drz#j۲Z]F̨@Z}RcĨq~4mND:dEYX=AoFֵW- w즗س`O<·?<2isP^ѣMV)𭩮4mqL~1bxƨ;U܁qO1Ĭ#a6pr5Si>S rph!=5:OB,ps8 /81lq0C3B ;>C :.C#MG;H*-m2}؋aܻ Ò!F=1,[ktDuu>iu]4"G5ƴv׬In]KZ;!Ib%[A"`arK+;@,PiÉ28љ wivt;~c.M#C KŮEjX﯂ @2\pOL:Z 5vvn7OENPY5h<Gah9Q~ΌSi{Z)v5}*s }bǛ ȤMP}ZFyHUMZ'ŀt) R' F%ɸY7:1Z `#U~"]{B6؆# &NrS.?+RՌ )G[Pf0MeZ1tfb6m["̓il)OYy֦6ɬhF#tN$pY{hNC)tޮ n+7BE9bGkO,Z+/p/ Dq[O(0-S"KDBEJ.YE5DvE6DN$w#bR7,&%OZ(Δ.h'3e0ȀO=(# vfH1Qeu}2oCU9AGi[a/-ep>Z )UX4JժpRr|ntڡNntZ 즃,UvX\Jލ ([SFғv6JT4}ޓ~Q)exõtc.BKz?NR+OaWi̓ 鑫Tһz~iuZ:F^KQsT&:wCM hK.,;.ԅ5k٧Nk~\Wy\ bNOZ#^ Hj+#fy`]=>;OF'2)̭kG׉FČO十d7`fT*17M=2=#3jQO+n%k , h!2b[/^AVasƂ=HMfxcwcn8e'wчz駊Px#eզčCT-ViV-HfhPXYOel-=+ZzU㋴/lZ\νhCP#PlvgȝYYqZnpCCz7#j'REc3> n?Q^p>/ӵ qW:L/[U~6σœ_ S.8QV5Y,hjd圫l[l ཿ:{L[ͶǑ!d; ^lߛiv*6K Mu&g_/ou w4˯[ fb}-anNOwߜK=C2tXg N㧧jbSsbc$02j#j9TX "Ə6(2:Q6ߦB.Uh\F's1I OKkb\BH=R41W\ayM){U''͐ɯ*\,GVf@ Uo$Wx@f*w+.u`= d0O}:ȗϹ:w<J, {f~UMwK[>`oV{OԨjJ5ѪS"bXcVΪlh3Yzѻ I'Nccu0$`f/@znU*.fvTHMΊ]s3qb^5ɸ[{:b8 ,豠MW';j{><$̈́'~13Gb\#Iy#=_?Xv t-_rfG}ŗDV./}|aH%YIˠݗuޮx m8hWЮBkѮCmHQh7݈6m5~6c:f17 l{q)?߇okk `N؍FprL*Zl;o/gh]r?ec шCш7%hĩhжBhĤhįh}NZE8VéϿ|wz=‹;ŗ{qy~ f;qsJtl1/8N=NYiBmsN/C~ L&e]lv?]﹧~#^\ AFH3} \)U,.C ư\CRðMth?7#<kBx k0Fq[Ο{=W̗X5Xr`՚M\܉;#sOڧn)SaaNa΍BIu;l?S,Dc,A#+AcU7Јadd<5_:!f_aRߴKN.^7YϲF+N̤ -ޮOA)jr`|EeC9 sIhw1^\7sS=!dJg^]hi0Dh}1=;` ʠ-g5'$%rzسPw8L|؆C=:;-~v{m=`iV%hUw}B>g[Ji}%EJqHوlu==ҺkKrNg_<:su^/ z ggaga0vy3`{uݳ,zaL0FyFMcjXSS#~ԈaLucaL3TH(rO~U斆á!r^RT+Lwf>玽QXW?G\iUR+EeF>eQ/rflֲzu2#u{X'3Yׄu2#%k:z4wdF>(̈i=7TaFbwѾh+Ի#1ugsxĬVW'Wq&bk#nz~Ioң<L,'}z: {<nJ&Uxh(:{G*IXôǽ#(ɤ\'<LNK&U!zǽ#ɪɤD #tT&'jeGDqHg gdR-J+,8ǽ#g%:gD9UFUc޹q>\%YL]XiuF:<:]VXo\;r[=ٲ*_[">9T[H݈ ş|JG.[V}rS9_.ETxA5V"OA9oHdRi|@GSpd7"Zʋ$K,Wߖ,-wQt&OtrN)GOGu.bRHdȂh4s,hf5Ts H8k)FMǵCaAίHN?,IOuAyVCٕճϭ?+\Wߩm 'L 5ئya@{`к:"^c)5NrQf-џHJ\z!xUQo\}lz' aSN km FEX u:ˬbX:P:T,h,^ڟ7>4SmΟcü,7Lm·e;fkyX;D@;eBPJ3ٙz^(P($1W l\w/~.Ȳf2ZM䳈fP5>e4SidcfC.T j ͐v ǯL(B!eᶟ,vnilG`Zj䲍GoAxӻ6^K,X]йTEMy<1eHfe=k3Z932 i.TRָ3e9yN >qgRO\=DNIYM|ۉ>DΈm<Ï-嚊=mQ}e1\څ 34$CunRd]ZbpK䇝OGgNY1#kF /J54mGhR#$s44?mDT)gm0DZTi.Hy5ӽ|G#g]tTGˎyeӡ!ɉ2oyURJ||H!pRъ/lGnI-~U&W\w w:Ua4<[LF $XqAr \}eQJvѵO'By=kVƛ^%Wǟo,~o[7+>AגOw9k\sүv^QS?7|i3n=X Sλf;cʌg;{}SWO{lYpOپx灜>θyz%w._a`ѶKY{TߋTrA_X5I'KJW?xkGfms)5~_~xΓ#~}ZO>説ӧ;ڡ(뽆;Y{/]noH'9vֻu}M{[OKoyˬÏ~w~{Ğ/DB}km侷C\mQ:gK$c!ˢYȆr cqԊŐ80>bԉ< 0ΥNq+֗'5?b5{[9PFY84& 5ۜ? [>GPgaVzvV ]u{9txYY>U=bWwgB.1/~̅~yjkg@hfj>BɏAO5Dz SB ŖJ5A Cū 2]^Q2(Kd?~^ PHa' `aGsb>ʄ2czaB+)b- hL#iOڌV `ZKUٙļ]''LTjskCQl*)iKv9# ?Qk*U;J2K81va M*'*"|2;ﱿcVlnp9p:sJ92s!g4aDވM;[fF[qvi3T}e]2Ձ/`ͭ nX% eby'i AJBg3?nrMFKIߚ-\JX'eS e>L M:'f/rRy2}7+WNЯlIҭE 7FLG*GtR^i} z:9]ɒf2&YQvh窃VzLa!|橃KI9J Y*@Il!rA,t1 ':W-*^*B DpYxIQ>U)GRD@<-R4*ҍ"wbX:*;V9Rj1'tղR=SG3|'=cD͛`"7]T(X|d{NBMN6(tPJʦ @ ` @@ @@@`@@@@`` `@`````` @` @` @` @`@@ @@@`@@@@@ @ @ @@ `@ @ @ @ @@@@ @@@@@`@@@@@@@@@`@` @`@@``@`@`@`@`@@ @@@`@@@@@@ @@@`@@@@@@ @@@`@@@@@@ @@@`@@@@ @` @ ` @@ @@@`@@@@`` `@`````` @` @` @` @` @` @ ` @@ @@@`@@@@`` `@`````` @` @` @`I. <eE aA i e eaabs Semisupervised Learning for Semistructured Biodata Analysis Hye-Sung Yoon, M.S.*4, Sang-Ho Lee, M.S., Ph.D.* , Ju Han Kim, M.D., Ph.D.4 * Dept. of Computer Science and Engineering, Ewha Womans University, Seoul, 120-750, comet@ewha.ac.kr, shlee@ewha.ac.kr 4 SNUBiomedical Informatics, Seoul National University college of Medicine, Seoul, 110-799, juhan@snu.ac.kr Abstract: Many supervised learning algorithms have been developed and extensively studied. But, in many practical learning problems including bioinformatics area, there is a small amount of labeled data along with a large pool of unlabeled data. Our goal in this paper is apply a new strategy in bioinformatics to semisupervised learning for classification tasks. The main idea of semisupervised learning algorithm is to consider the problem of using a large unlabeled sample to boost the performance of a learning algorithm when only a small set of labeled examples is available. We also applied a co-training strategy for unlabeled data to improve the performance of standard supervised learning algorithms. Some empirical results for the suggested co-training strategy on real biopathway data are provided. Keyword: semisupervised learning, semistructured data, XML> w DM(Direct mail) ewa, aw e i i e a wi xe A { baСa ae e(target value)a e Aa e bsi aabs(supervised learning)a ae ee bÁ w Ae ea ae i aabs(unsupervised learning)a ea. e bswe a qae , AAa A, A, wi qa A aa aaA wEa. eс b(bioinformatics)A a(bio) Ai Awe i 贡aae ea g A a. aСe b ae wѢ wa e e AiA, xa ae A sE ʼne i ‰aС ᶁe a Aȁ wѢA e e se w iqi aaa. ᔁa bA aСa ae ge Aȁ wѢ eE A(semistructured data) i a a. aСe e A A se wi x i waa Awe iti ᴁ ᶁe aa eС i awi g a. aa Ae e i aaa eE a AA ia i a⸥ wi wae wA eaС i wava. A b A eE A wѢ e XML Ai, wae eaabs(semisupervised learning) w co-training iqi wava. a i Baa A a aa wA eaabs w󤡷 a Aȁ wA b aa qi iei a. _