Scribd
Upload
106 views

Soft Computing Notes

Uploaded by

RITIKA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
106 views

Soft Computing Notes

Uploaded by

RITIKA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 42
Soft Compuhing eel® he. aCombudin = : Antecedint (x) —2P FoQ FY Conneaut ings ox am eve Hh {ins ran cnt easy {fouled} Nahin At Praja of ConfahrgSyahn: tatty eden Rracisa Solalion Contd aulon Unonebiguons 7 Accunade a Matfmlical MactL Ty he of Compulng:— Hand Oompahing Hand Combubing + Sof Computing, D Ravull mat be Pacer Hy Laid Comply (Quant 0 oF) ppaon-x Solt Compuhing.: @ Contael Aelion must be © Tope Sa : Form ; one Sw Numunical Rasblom Um cortomi same ® Saanching. Probl a y Sameer © Soaking "abe @ Low Sobalion Cost ore eer © Deo not sepi d Mabhums cot modi | yalalionany [Ganaki Alien Fens, Logit 5 Newel Njo roltionany [| bt tig aod = Boctr Ve P Seo exctte Roll Lonel on é oe fsione cla : ae a Hy laid Coomfpubiong, alan Sobveddy 66 -~ io So afhse US] Hybid e Rracisiom Tmbncision Bared 6 din Cased on Fi ' wei 7 Names so bl degic,* a. Neocles Seedt Tela, Deab, with amlinn 7 nee Sbictly Sequombl Allous- prroLbeD Compubolh Coubainly. al bin 6 ) oa Seay! + Ml fate Clee lay) brecers the Ip il Net Lp Dendrite: long inyadanly. shohedl fulamands ablabled te He Some - hp isha Axon Prob Ys dink ae f th wm omare Pps channel | ofp thet cantires the dompulres Gulp of Fhe Axon? Vablage os ea Schumardit Drigram of Biodngcal Newren > Synapre Negcliws oer Ni Somme, < Ww call Jody. Daondnites Coane u spy Metloubeal Madd of axtificiel Neuron Cc f oO ¥ y oS ee eotbet CS AN oo, Chornunss on tech Seat aan tpecified ig teu desi. ember: © Commechions > Single - boyan seb fennel ne 3 i, dhe pe ea sath ale onan jpotbac > Ginglebog acum nfo Feb pho lage sent Nf Srbuurid Lusaming. wb bb tW ejay Gipaath Sipmsie 4 Unoewbauiied Hanns wo Rainfrvemment Learn, re ria Side Signi 4 @ Single node with own dfecd-back No Ad vhow OD Tebolily Ys Linn JP feo=x Tin Yes care. 0/p Jumains bare. os T/p on DP Talundity achivelion pi @ Binay Sob fel: Joo = {4 Jno wf — 6 J 2<0 6 he 0 Honbold valu <>} % Gonnably waa im Aingl~ Lagan. Wa) do Comuh the No &/p to o/p in inary foun- w @© Bibebor Sts 2: rs) wus! Hxr6 oles at — 4 {\ yg eile this dis ured Jin gringh— dye Wo te come the io | wd x/p to ofp sin dipole (41 00 -D) i Be sulolionshib Le™ @® Supoida) fulbs— Used im hack prepagolion jg Hecawss of He ral of Ha fol of a fo? Hf veloe.of Hecho te Reduce He comastonal Auncln alvving’ taainseg jgmoidel Jos tyes : ee 1 jg Steines parameden 0 Binary Stymie we {oe = ae Ung Soe 19° sar of tobe th foo abe 4) Unisex Samad 149° “sanyo signi BP fom onl; © hel pie! | ae ant Te i sang of Hin fob AS™ —1 41 Deira oie obve He fered pe) 4 © Ramp ol:- fof ae © apx] =[o8,0°%6 on4] [1,2 J =[oy 03 762] sas = 073° lecaae th yp olueged Yas bt Be “ Yin = O83 (Dd z 1 Sal = 062- J=4Oun) 4 ete e038 r 2 =o. @ y= flim) = wr —1=0:259 = 2, Linpbrand AND fol unig MP mewron * x. ; ee y the onby amas i 1 ise dake weights oy =) 7 a=) | ° 6 i (W Sip = 200) a0n= i HM 22 ° 5 (LO Yan =) ° io (01) Yn et (40) Yy=0 Yin @ ot CHD : : ® Yin = 4 ot Lip i awa fol? %p » Ty A dh te Tpprore hy). el ot (on Net infu = 2 for C1) Sp Aosedd on thus vada. net x/p voll Hassell .is seh Yinzo at (0) 4 Hans if He Till val 40. > gual Hom en equal le 2 Yun raucen frss- oP aewis net is Se (e= 2) O=2 abe Ae pod Ay @ >nW-P ne2 wel po O22-0= OF2 So ofp of newenY Can he written 0s ye dlign=f! Fine? oh Yin <2 fap: 3 Limplemonit XOR ful” ung Melulloch-Pitts neuron Sob TT of xor my Ha Sng Augie fo iY 2B + 1 2 ys ztZ2 wh zeam%m (wy Zaz Ha (p42) yeztZe (429) that means Aingl degen f= nehsock. ax nob aubficuard ty cupresent the oA Se na al byes pees ¥ 4 . Xor canned be, aspensenbeel dy ainply i ° Cort Pyum Wy, =! Wa =! z o nbz/P 0 Zin = oxi rom =O 7 CV Za =1 U0) Zan =I e Ch) Zay =2 Hone E+ is ned powibl fy full z, using Hse weight Gone y=] Wy=-l fea wt E/p (0,0) - Zinz OXI tO¥-1 =O |O Ox > Cond (0,1) zmzonl tix =AF\0 ea ~ Ub zm eter tox! = Th (ol) zin = a +181 =O IO S® gpnwtP pyar or! > often) -{! Bind! O 2Zn! for 3? fut y= 2, +Zo at, Ls =. =o, ee Rasim Vi=l Vel Ore er Yin =Z1Vy + 20M, ' if F, 1 (0.0) Yn =O NI TOMI=O Oe meee Cop Wa cont tte 5 | 40 G9 Yin = rep tow) =) y GUY = onr+orl=O Vel Linea Separability - @l bru Separability as the comcept asfencin the Atpotahan of He Z/p space into aaglons Habis daseel on wfuther the nfo respomse is posilsre on nigabive» © | Deciwien Line is clrawn te suparek positioe nugahic cxaporse} ©| Deciven Line 02 clesnisn-mabring Lins on oleision- seppodt Line # Hum agheodl baa To chasify the halhon Linear Sepnebihty camcephip used te Conceppis hered! om output Rupense Wp dapat Samm b+ $00 Ol Respanse Bored Tf Aipabr beh achrahen fal? a weal over the caleuDaled mut x/p (yy,) the value of He fo™ ds 1 4orvie 7 -1 for —ine T/p- hed: unm Ha ane Aoundug 4” Y,>° % JinSO So cher'sren Aounclorg obetumine dy. Ha selon et Zvi =o Far twe X/p A+ Wy2t) + Wa %2= 0 Nebwook ru Recsrcing 40 Hebd mu “a de weight yeotor is founel Jo Anusaye proporbionatedy Yo He produc} of te zh SF Loaxming signal: Meaning Signal meant He rewron's o/p height eelede any HAM, Rule {Cay = Wj (ol!) + xy} = Hebb uh ia mare atuided Jor Lipoler clad Ham inary, dodo + Lf brnagy chde ss used thon weipht updehon formule Canmat obishingunel two camcbihon namely - @ then Abrorrntsg pasr sn which am x/p ami bo ft © when A troirvig tn sn ech ot do unit hay Yel Toning pow thn— ohh 9 asta He weil tegdl mag be tuo: 4:6: Wi=0 for det ton whee n= Tatol No of T/p nusw0y Shp2: Shp ates tare tote perfec for Gok ligne eye bi \ Sit Sb 3: Tp units achrasons are set Ganeredly achrahon Jo 0h ofp.bayir 0 donb ly ult =; fadtziton Stab4: ofp: unit selpetine au aut Y= Ships: weight of Bias acljustmnint-are onfororedl: 00s (ne) = Od; (old) +9 ep (news) = bold) +Y Hull. Rue is uted for pahteon chesieabse , liyseizeden if ansehen: YY , . a Gah, Deum a Hebb. net ts imp donot gical AD FP? (se Aipalen £/p of tengeds) m4 x + Y Th ryfo ib raimeeh wring Helle fo training. olgpiften Trikially cacights g bias ath ty zen W,=0 Wy= 0 4=0 Fe the Fish £/p [x 02 b= [11 1] F tayél= | (ona0s 9=1) wp Uwe) = 0} (eld) 365 y (nur) = 0, (ory Boral =! WCrus) = OzColl) tay Pot IXl =] pp bows) = +l) +9 3otl =] Foo Seaind Z/p Ligh 1] tug =o (ald) rary 1+ Ie =O , (nas) = Y) ! An-l=2 (nut) =uplold) +2aY SI + An) =4leld+y >! +p =? Feu Thixd Z/p [1 ta)) tay =H! 0 (ww) = w, (old tay) @O FT x1 =! walrasy= enor any 22 +61 =! weh(ws) = doldty sorta For folk peta feagel =-1 wares) = ela) ry BL PD = Z ; Wp( nus) = Walold) +92¥> | + (--) =2 Fra la ts jue) = bold ty > 140) =? L, ie ek y wily Wr) 20 eel 3 : oh eh i 220 2 ieee Pas insted oh 6 Fax seperoing dre eft ay given dy ah = fe 7 ae if oundeeg, divs 1h” thy valuos 2 20/940 hel rah girth de tire vebhonue on one Arde ¥ ait putponse am Ale siden th gia as Ant C/I) +2g = 0 4a- O ye oy sh wad Fou rrp Dit) wer ur esq! wan D> wy = 1-1 —D fu x" Up fad ee. a=2 bao AqQ=OXX)—6 => %=O —0 Fou mrp [iy 1] wet yeas b= Al y= she aoe aye yt! —@) Fx r/p Etj-1 1] 22 meee ph=-d ag ae ha ES eq = 41 #1 ——D a a oo) qe oe] aly bgt BP | ums + xed x, ; al = hy y-| Cay ae i - bd oy Gn iy (a oy xt T/p Ran x xp mew r/p (Douision easy fox AD J ning Hubb gulb be cack buryg bus) ay Phen Neo ett co) | Unit (I/puut) @ Noseclader umit ( Hidden unit) © Rusponse Unit Co/e unit) ofp (0/1) ofp Doind ised te oft o/P sd ° ea x\xfx <| See Lad @-o* Susi @ ®@ Ym tn Somme gd — Rasactabon- ‘ shel unt Ree ; © Seuoy uni, connted to arscieror arts wth forad nigh Iaucrg vodbss \y 0,1 ice is assignuc) of wxamcom * © Brracy achialon fl Bs wel in eta antl and ctasciche unit fame unit Basan exbrelon of t01-14 fle inary sip itl ed tube! 0 is wud or FE hralon qo pscahe “ge us tab muons the ofp signal Ha axe seat from avsecrahion Welt Pe aurponse unit ane areby binary: © ofp of He prclaon no 2» y= -fsy) wa Chen) = Meiehon fe I ah n> dim |5 4-aege oy af Yn<9 eet ae au ofp for weight udelion el’ asciae ¢rasnse © Pun coleulhre ssluudl on He camasirom of ts values of faagebs ont ps @ % waights an He connections feom He units Hat sere] the nwnrer wll gd djs satel # eB © weights ail apo om he Ass of naming oul ef azn: Aas ox for 0. prcticulan tralning pattern co, (tans) = Cay old) pet; ‘: i plus) doll) +24 { wd omg ole Huo 4 16 welt apdehen J hance the Peorming xf No er2s2 pcctles , Flas Chad for Singh oh Persbben Wo procs hp AND i nN) : (ametigad eas a ee ne orl, YxQeb=0 ¢O=0 x21 ,oco-l 2 f re 2 @__wizo le w (ns) = O + 1X1 XL = | 4 : ies —>y\ w,(nns) =o + lyr) ¥1 AQm sotlet =~ Te Refoons = A lon ve complete all dexp Gud) Ba = bedtorns Hun 4 ebeh epoch comple. YWo% 1 tla +en7e) = Fu DM rfp HL] eta) - Se! Neb L/e -Yag= Ae Pas t%2 a tes ee 6 i) ra jin=O 1 ud — “ge Xt) aleulole voy ded 44 =e oO { jin = A th igXO > Yineo ms lin Xi Chuck wrhtle toy Ho t=1 PIr? S t#y Heo. tees the weight es waaay = (ald) 72 = erininn wy (nus) = o+ 1x1 * =| J (ns) = ot Xl =) bys | pedal gba Fa rp =o <7 Yin= + iat 4 se y=flad=! chek te-1 ye) > FY) incre) = Leet On) Spt led = 2 0 elo) 202 Eafe Cit t=! Yn = OF ORT TE 46 y=4Un)= | t+#7 Cit) agen Upelade Wx 2x) =2 Og) Find he nigh 7m ting, poicapbron ric fr Plosing fotteors- ron all He ppsndre) only. one Hine - (sL=1 8=0) CO Grpabs) weg aa x. Yn wy eb e008 1 1 -) ° oa oe \ a 1 -t ~ Oo -20 a ; =! Zt o -20 ay “4 a 2 I poe! yee \ Find He peghe suypind Jo pro te following clifton weg hon nfo off yectns Url, Io 1) ane) Cy ty-tsl) ebb beongng te es 2 Ges d and yectos Giotto) and CL“ lene ot Lelongirg ie chose Choget=) Panui, Lemons ake ot ood smitjol Ma of egr=0 IF Epoch eee a) i tog 2b = (010 1010) (oat! 1 ° ° tod 1 ital a-' a4 ) I -I Owe DO 2 fe alee eae el ee 4 ! 4p vest oll 1 -) cl tal ! | 22 0090 0 Eboch~ ; ! 0 ° -1i9,1,1 1 le 3 ! -1 3 1 1 yn wd [ies | 22 O20 -| - =] ~ 20 Epoch-3 1-2 ) 2 2 10 von on» p22! -2.2 0 2° 7 Zz J —2 2 0 20 oe -2 2 020 —-22 02° -1 -2 I Adaptive Hineen, Neen (Adaline) /nyo @ ee peed Tn peecepdren Nita Hearnés caulk dase en Heldian Reb But Im Adaline Learmeng Rule cis Lesed om Della Ru © Downed om, Gunacliond- 7 do Adjusting the cocights of iA potbern (s'=1 ton) he my cl ele ar Wig este oak ¢ iy) ge \ anne woe : ny a opoun |e Seal ofr et oye Baie, AW = & (4y ~ Yj) \ Jor Sucnel ofp units Yin aut Sos Fog }-— Yio Tabne|e_e=tt- Yan) ; aie [esto prearE —> f Turing. Algorithm <> Stcbp— cond bios volwe un grande bad nod cere y et Le thie ore a Step 2 Penfoum a fo 7 when tobi candidien issfebe Sttha Baform aif t-te 6 fer each icpalee tara pair Sx} bib rt So acadions fv Fp uni, d=1 don ej = 8} Sthe Cabeslale the wih yp etl ofP une Yin = = bt SW; ® skbe Utpaete the cocighh-9f dies de iziten 4 (nud) = 0) (600) + O-9 i p(n) = eel) + Ct-I7'n) Qkby Ty pighut coer change tad ocountedd alating aang ie 7 =H ited ei Ialerance thon top He tromns pseu ele confi nue (Stehfeng. Condition of We) Topo? of -parcher wth Life ps and targa orm elas. x 2 + Iotot | wield As cabubled unt fe esc &, a a) i a - 7 sol My cw, = bro! f tao! Rafe gol in act begs CHAPTER 3 / SUPERVISED EARNING Nt : ~SWWoy 12 ks are assumed t0 be stall nindow valygg ek here the Fes mean Sauer gua Oy an square error is obtained, ‘ON = 0.1 and the les we calculate the 4 Illy al ee eit a the leaming rate isalso sot 00, weights are calculated The initial weights 3 eae For the tet input sumple, x1 = be obs ran ial Jin = + xy wy + XW yin =O1+1 x 0141 0.1 Yin = 0.3 Now compute (t ~ yin) = (1 ~ 03) = 0.7. Updating the weights we obtain, w;(new) = w,(old) + a(t ~ yin) x; where a(t — y4n)x; is called as weight change Aw;. The new weigt rng ey tea Net Input ay “OL x hts are obtained as wy (new) = wy (old) + Aun 0.1401 % 0.7 x 1 L+0.1% 0.7 x 1 =0.17 b(new) = b(old) + Ab =0140.1 0.7 b(new) = 0.17 where Avr = alt — yin) Aw = a(t ~ 549) x2 Ab = a(t — y,,) Now we calculate the error: Es(-y,) = (0.7) = 0.49 ing first in The final weights after Present Put sampl le are w=([01 and enor E = 049 10.17 0.17 0.17) E + 103 The above calculations are performed for all the input samples and the error is calcu- lated. One epoch is completed when all the input patterns are presented. Summing up all the errors obtained for each input sample during one epoch will give the total mean square error of that epoch. The network training is continued until this error is minimized toa very small value. Adopting the method above, the network training is done for OR function using ‘Adaline network and is tabulated below for @ = 0.1. Weights Net i ee rarer ace Weight changes =e ee nal t Ya tin Aw, Aw Ab (0.10.1 0.1) (tin)? EPOCH rit ot 03 07 007 007 «007 0.17 0.17 0.17049 1-11 10.17 0.830.083 -0.083 0.083 0.253 0087 0.253 0.69 111 1 0087 0.913 0.0913 0.0913 0.0913 0.1617 0.1783 03443 0.83

You might also like