—BACK- PROPAGATION NETWORK *- a 2 Tt_tb pre of the most important cleselopments in neural russ 4 The network has awakened the Scien life and _engineenng _ —____Conamunity te the modelling and_ processing. af mumeseus quanktatuse phenomena _ustng “nensal nebeorks a \_b Tt is _applied_to multilayer peed= forward networks consistixe bf — processing elements totth continous clipterentiable ackivaliin — eeu 2 BPN teasing algorithm ___Date MDG A CO ODE ELE LO LL LE é provides a procedure ~forthaning,. = the vols 1 BPN to Classify the given input _pattesng " Orr =— = —Basit_concept por this, update_algorithm_is._simply the Gradient -descent_methodas_used th case ef simple percepban metworks with differentiable units, — ‘s—This_ts_a_method. where the error_is propagated back ta the hidcten unit = | : Ti_is_dijperent other networks _in_resp cess by | ——twhich—the—weightsare calculated custng tearing period __ of the jus. by General cligpiculhy waith nautti layer, perceptrens (6 eakeutatir the wits ef the hidden layers th _an_es/cien| gay that weld —nesult tn_a_vi rere _Outbut error. when the hiddur = —loyers inereased_, the mjuo_trai i Bmore complex» ——To_update weight, the error must be calustated. BN ie done A Segui _feed~ forward of the fnput_training pattern __ * _Labeutation. ane back - propegaticn of error, *Updatiin of votighte Scanned with CamScannerfia: Avchikecliiee of back. propagolon nfud Trash Lina ug Algorithm en Step Ot Trikiatze weights and learning rate (tak. Ta , 2 a (take some matt Stept 2 Pextorm steps 2-9 when —Stnpping conditten us false, —Sep2£ Perform steps 3-8 shor each training ee am Feed fowsasd. phase ( Phase Z) fs . ° + i. = )Step3i—Each_input unit seceives faput Signal 2° and sends tl = tothe _hideen_unit (f= I ton) Stepy & Each hidden _unck 27 ¢ f= Ito p) Sums ii eu a to_Cabcutate net inp fej t Say vi Scanned with CamScannerDate Calculate output ag the_hieleen unit by applying fgmnoictal_act a ——funetiéns oven Einj. Cbthary or bipolar _ fae ef zing) | Bang send the output signal — fromthe PR put of output layer..telles —_—____ ~ gs Step 5: For each_owtbut_unit_y,(_K=/ tom), Cabestats the nek i[p == es | gio ok + o% Ler K. and apply, the ativan _sjuncliinta_comtpute_output_signal a Phase ID) :- ep& 2 Each output unit Yq (K=| tom) recess a target ing to the iny Enis on ¢ ~ ewes Scanned with CamScannerperivatise 9 binary esty moet ‘ pooe fr C1- foo] — Shep TS cneh_netetn_ an sprom. the output nile [r= Gini a » cateulated cle epenching on “ahether Te derteabios pL binary _o1_bip' iu_t_used - nthe bast 6 ok __Cafulated 6F oak the change mn_wls_and bias ¢ Stepan man pt lk Ce K eg ee 1. (new) = WyKCotd) + Awyn) wo0K old) +4 AWwok Scanned with CamScannerperrvatiie of bipolar Sigraatd yiws 2 ed ‘exporm Steps 2-4 forts each ipput_ )__Step2.s Sekt the _achivation of tnput_unik of 0% —Step3t Caboulate the net input to hidden uncle % £ iy output Fos. f= | te p—___ aac s ES Step4 & Now Conapute the outbut of the eat yen nik Fo, K = | tm, [rite — eg 9? Using back Propagation _nebwork, ind. the new wughts pes. pet Shown in di ge pemeniee eath apu patetn EO i and. ek _output is Use a Leaning rate #20:25 and _ = a a funckuin- bias 2) et O38 piaegrey SEF ; -OL Nig = = 0:3 Vane 0:1 _ Na2= O'4 Voy =_9°3 No2 = 0S" Scanned with CamScanner__Amput- Sample [x1 _ aa]: 2 Le 1] target t24 1. —_ ~AColeutate the nek input for hid aeen_loy ee. — Fox Zy 2 I Zina = Noy + %1Vir t X2Vau | =_0-.3+ ox0-4641x-0'l ' = 0:34 06-011 = 08 Fa, Z = waz = Vort %vi9 + 2 Vor = O5+ ox-0-3 + J XO-4 = O-5¢OtO £09 a a 1 a = 0. 5498 =f (Zit) ree ET ae ee = 110 Za = flii= ee ee? “ Scanned with CamScannerig ee oe a "C4 = iy eID Blqut) a 7 FE tear) Cyn) —=— 02495 6,2 (1-0-5221) (0.2495) = ONL ‘A Find the changes ti weights between hieteln_ancl_ouspuk layers Aw) = ¢ 6%, © 0-25 X0:119) ¥ 0-5498 = 0-016Yy Aw = «829 = O'R5X ONG ¥ 0-709 = 002 t Ab = 46, = 0:95 xX ong * =__0:02978 A Comput the ener portin 6 behoeen input_and_htdldin loyer (2 Ito 2) ops Suis’ [Ziep) = = Ga; = = Cx Wik Here Kz=J only one output K=I neuton Cunt = Gy wy, = ONL XO-U = O-pUTEY — bug 2 By way-= 0-9 X 0-1 = 00119 Scanned with CamScanner——Enor_Gi= Gin (a1) — _t ie — plas) = g(2air) Cr Peace] = =_ 0.5498 [l- 0.5498] =O. 2015 a "(za L) — 0.04764 X 02475 = 0-018 on f(Zusa) +h (eirg)-Cl= 42a) 0+ 1109 [ t= 0-T109J__ 0 8055 G2 = Cina (Zin) =_O0-01195 X 0:2055 = 0.00245 AAous changes tn weights between input and hidden layer’ —Avy = & 6x, = 0-25 x0-0118 XO =O Ag, = 4 Ey 4 = 0:25 0-0118 XI 0:00295 ANo= x By = _O-25X0-0N8 = 000295 Avin. = 4E.%) = 0:25 X 0-00QU5X0 =O A N92 2% 6 %= 0.25% 000245 X1= 0-0006125 ANog= 69 = 0:25 %0-00245 = 0-0006125 lompute thefinol_weights of the network? — “ _Miylnew) = Vy Cold) + Av) = 0-6 + 0 0-6 Mia (new) = Vio (Old) + Avo = - 0-34 0 = -0-3 Jor (mew) = Nai(olel) + Ava, = 7 0+! + 0.00295= -0-0910- _Vga (new) = V22 (old) + Ao = 0:4 + 0:0006125= 0-4ooe1as Vo, (new) = Noi(old) + ANoy = ©-3+ 0-00295 = 0:30246 Noa (nay) = Vo2(old) + Avor = 0:54 00006126 = O.Soo gas Scanned with CamScanner’ WL (neue) = wy, (old) + Aw = 0 Wa mew) = We (old) + Aw2 = 0+] + 0-02/1F b(nuo)= bj oid) + Ab = -O-R+ 0.02970 =O0:11022. tad_the ne » weights, ing back - propagates The_network._is_presented with ‘the Input —pattsn [=I 1 —And target tb Ls Use | ng_rote 120-25 ¢ byereors Vi20:6 5 Win 2-0-3, Yap = 7 Ol | Vag 0-4, Vol = 0:3, Yogz0-S Wy =ZO-4, Wy =O) , b= Od pat Bipolar _sigmoteal function + _ pis 2 -L = b i-e™ ie I+e7* & Coleulat, nek input spor hidden layer _ For xs = iin = Vor + 2yNu FX Vay = _OAtENX 0-6 + 1X (-0-1) = 0:3-0-6-0-1 = -O-4 Scanned with CamScannerNog + %) Vay + % Vag tt o Gite sf=t) Xe(2024) fe eX Ot 4 OB +O 3404 = 9.0 =O 2 4-0:1974XO:4 + 0.537 X Ol — 0:2 +(-0.07996)+ 0-0537 eb ege see == 0-4122. $bwi) = 0-5[ ln O-J122 J [) + 011224 = 0.4434 Gre Clp 61122) (0.4934) =| 1122 X 9-4937 = 0.549) (Find the weights between hidden and L output layer Aw, = 06 8% = 0:25 X 0-544 X_ (-0-1914) = — OOH Ag = 462232 025 KO. 544 X 0+ 534 = = O07 3) ey ae en Ee ae 20-1373 - Scanned with CamScannerce udp. p (Zing) — _( there 1S only, ons outfall PGi 2s Gaal 7 want = 9:81864 — ase L = 00549 X0-5(1— 0+ 5371 (l+ 0-537] = 0-0195° i ; " . pA Changes ui weights between inpuk and hidden layer, ¢ oas Xoogsxl = 0:0049 O:&SX o.019S = 00049 ki Avy = x6) 21 = _0-25X 0-(os6 X-| = ~0'026Y r ANay = & 6) 9 = Oras X O-locex| = 0-0264 b 0. as X O-1056 = 00264 b Oae X O-019S K-| = —6-0044 b Y Vu(nus)2 Vutoid) + AV = 0-6-0264 = O- 6136 YNi2 (mew) ANin_ =O: - 0-304 | a a) =Nqlald) + Aes = = Ol emsnas 2 = 0-0736 yNa2(nem) = Voa(old) + Aveo = o-4 + o-p049= O-4049 ae (re) 2 Sigg Bea, 3 0- — 80211 5.37.24 —- iil 2 Cas et he 2 = OF 007131 2 01737 st “gs, Scanned with CamScannerYo2(old) + AVo2 = 0:54 0.0049 blotd) + Ab= -0-2 +0-1373= = 00624 ASSOCIATIVE MEMORY NETWORKS ___ly Tt _can_Stpre_a_set_o4 pattems _as muncories. le _the associative mma is_being presented _useth, O- __ _Key_pattun , it responds by preducing ene of the Stored ____patterns,_with—the—belp of—infermt fin —mremerized which ___tlesely, reser bles—errelates to _ths_Key path Thus, the recat” is through essociatnn of the Keypatlon ite Wa aa | ep of —ernformatecn—mtenae! ae bs These ty tt » also called as content — addressable mentersis CAM) th_contrastte that of Lraditimal addrecs- addressable mentories_ch_ xy) then the hamming, olt’staunce ( HD) {Gh doped _as the ne. of mismatched _Contpenents of pss zand x’ vectors ie. 7 7 as Ne Zo faye Xi] if ti LOU Lo ip (ayes St ’ cee pee _-—____--— a = 2 Ly Tf each of the output verters & .same_a8 the input vectors |” ___ with whith itis _assocfated , then the nel_ts Said. te by Ip each of he_eulpat veces Ore different yn the input vectors then the net _& ssaiel to be heleroasocvatise eusal nek. are updated until there is no wera ht change Ly no eughis Sep O: Set att the initial wlt te zero fe. wy=0 tapas ad Steph! For cach tring target pat apt ea eyo Scanned with CamScannerSte 22 Aolivate inp oye unit jp ewrent current ig — inputs : - 5 meer | C4er- (2 Iten). -€ Stop3 3 Ackbati the output layer units to_eustent target — Output 7 z ss Steps Start the wer wif (new) = wis Cotd) + xy Thi algo #_usecl for cateulahin of the weight of assoctatise n le Tk tan be used useth pattans that are being rep ented As eithes binary or_btpolar vectors Proolucts Rule t $- i oo oe Tnput = $=( si, +-- St, ---Sn). Pe Outer, product of the two vectors is the _procuck of matrices S267 ond T=+ i-¢. [nxt] _+[) Xm] matrix by Matrix Mulbplicatiin fs dene a8 fotlowss $= : ST = Scanned with CamScannerDate ecght- Tiare Us same a8 the vw —by Hebb rule to store the pattern For oe a Set_op association S(p)? Lip) pil Pookie a i ne ssk_p)) wet ie Et pei lpy by (p)-= ta Ly) peed rece vip _can be quien. a Scanned with CamScannerys Arabians Trang Alger than? Step! Os Tnitialize : ol sought te #8 wits Le Step) i For each op the vector shat hes te be Slorad perf 24 pitecture =4A ‘aha — L_step2. t _Adkivatsenth_ effin sbjuk_ unt att aps Ac ree each p i i. | Step Yt Adjust tbs weighs = : __ wife) = wyj (old) + igs at 2 Step OF Set the wls obtained or Hebb rule or outer products oe 2 For each 4 testi )_tnpuk vectors pererm Step 2-Y - w=: 2 ep) 4 Cp) Cafcutate + the oukbuk by appli g—acbivaleon_ower._ net f = Hy bly) = ha Scanned with CamScannerZ —— a “ used th speech jrnteig. sification pet. Ly Thi +: ype of metwork tan be 8 processing» pattern elas taTIVE NETWORK $= ___ 4 _trathés input Pal target “output _weetors rs are agers __ 45 weights are _dleterntineel _th_a_way_4 1 that the net Can __ Store a Set of, pattern __a.sso Ee by Assotdakion ta pais. a tng - ___paiss_{ fp), 4p)) sith _p=t,- -.,P_» Each vector S(p)__ _____has_n__Components and veetes 1p) has mm. Cor Cont poneuts bs _peterminaligin of ust clone _etther by lng Hebe th — delta rule. saree van appeapeiaBs output car, whith cores pads. ie vector x, that bey be siete one sf the - unknown input. [Step Of Taulialise thes wits from the trains | SlepL? Peja stp tren inp at pose ‘ {—_--- ics a Scanned with CamScannerData. —Step2: Set the ackivatun -for_input layer units equal stoi | that of the current input vector. guiin x _ 7 3+ Catewtote the net _thput te the output uviils + — ees — fj Step¥ = beterméxe the activations of the output nits over —_<{ the colubated net input 7 -—h pe SEES pigs th yur j £0 NOTES Heteraasacitite mtmory_not_an iberatiise memory netioork. Ty the respones of the net ore binary. thin the ———achivahén funttuin fs be used cs a Ye 1 yout) 7% O os f ying £0 __ BIDIRECTIONAL _AS9clATIVE._NETWORK b Te was developed by Koso is 4he_yeor LIBR. BAM nju performs forward and backward assoufatuit searches fo Stored ehimubus respprees. % Tk l_d_recument__heteroassoctatiitfatfeun -_matchiag, —nekwork : ole = 5 BAM _newnal nets are capable te _eepend. “te shbuk Pram eithes. _Cfnpout_§ output layer) Ly There_ore 2 _4ypes_of 6am t @ Discrete ee eee OE ae Scanned with CamScannerE _biserete Bidirectfonal Acsomatuie Memory <— 1» Structuxe of oliserete BAM us ame _as about. _ Ly when the mene are_being activated by, bg a @n_imitial vector at the input of 7 Layer, the nelson evolues_a two-pathean Stable state wlth each pottern_at Output j—_Lek_the iaput veclor be denoted by sCp) and. ta clOrs_b ACp)—p=t,- P._ Then the wot - matin th store a set” of a xeckors , a | S(P)= (Sip), 22 SUP) yoasa Sal PY = Scanned with CamScannerttp)= (4p). .-= . tCp)s.e-atm-(P))— —Can_be_ p Biteiiaid by, Hebb scale train 0. In Case of tiput J Nectors eg binary , the wt matrix We $.0ij} i guren_by, £2 setpy=!T a Op the _aiha_ | aahen 4 the oe me vedi ae prt maine W=> {wij} an be defi : a _ wij = = Sip) ee Cp mata in both cases is go be in bipolar 5 efHun the [nput Veclors are in bisa Scanned with CamScannervoit bra fase hueedeetneeta We L mn ¢ 2 Ot mi je i ani = Of eee Hg wt 2 OF Teshng Algorithm Jor Discrete BAM,£- v —F 2 Pesed en the trati Op wes ced ©, i ab culate: 2. acti} & a recogutze the Diogal _ 2 __step4s Ls Update the ~aitisaliont of unit in. Y tayes Catastati the net input ayy 2. ase Sétnaé Scanned with CamScanner—obtata 7 yb Cys) = cast, wo Updali the activations of units th X layer. Cabealats the net input, Apply thi actisaliéns pyen thy net input, 7 fend_this stanal to the lay Shep 62 Test pos Comvetgence ef the met. The cemueigente occurs the actusalicn vectors x ancl _y reach equilibstum Tk _transjovens the infu ethly and. comirously ha Ons using legisltt cegmiid pundiias as the Aclwalion puncions gor att leit, —Logisti: sigmoidal punchin ring be etthus. binary signoidal ——fundien_es bipolar sigmeidel punch —'_when_a_bipolan sigmoidal funch ———thosen.,then the Continous Bam might ¢ Of —vecllirs_mhich will afmath vestiies of the cube hin 4hat Slate of the vector appmachus it acti like Tg the input_vectors ore inary. Le(5), t(p)) p= foe —weights_are_detenmind using the sposowuba se lasipl C2ti(p] —_—— aes wif Scanned with CamScannerbe. even _ mee the “inp ae ae ut th -matnia_ b Ee fo bipotas. a —— - 1 Actiahon puostin. pol Legis sipcidal junction a ik l4_bunary logistic. mee »—then_achivatiin tauacbuin fo | i+e eg J i_a_bipalar, loguslie function apse fe et | [+ ed i eve | Scanned with CamScannerJiscreks. tdd_netivark —_— 45 Tt i an ute asoniabie tatereennnted — single : metusork . Tt is algo a ype nebrically_sctighted nfo. “9 shen thes Us operated (n_ctéscrati line fashion itu tatled as : ~Aisercte Hopfield network § A= archickective as _a_ single slayer ‘- feedback metwork _e/d _yeeurent nyo. —b__Netwwerk takes bihan L) or _bipolas.| : yl, P Nid has Sypancbieot li ttl nto self =cennsttibest yf . Lope uate twin e0. —— Key podaly = Hoon et tril” updates ils _achivatiins at a bint. — Lo Each unit te eonting sly erosive extern cxnal_signal al _ sills the Signal _it_receives frrm_the_othen unite us the . met. le layer restient_in_peyeening depute Luptatig t_patlesn is just applied te the metusrk anc the ———tashoork's output is _found tobe _initateect accerding ly 2 4 Aftesasarals, ihe inilaliing —patizsn_e_remosel, ane oxttpat yy — $_ initiated betemes the neo _uptated {input thnssigh feedback | —Cennestions. ‘The irs updated fp forces the fisst update dt ———dutpuk, whieh in_tusa__acts_as the second —tpdated_ input ———thaough —petclloack connetbitias _ And tres cult Oa _b ths Ail be Continues till _systom_systin reaches a tin_secend mie Scanned with CamScanner| _Output rem each —proeecsing : pup ctor process )—elerseusti, but 26 S{p) = CSL) + Salp). ze [2syp) 1] (2sytp)=!J gor C8) pal Loughe matrix We For pales tinal geet = Site) Sp) _#f- ga = Scanned with CamScanner~—Tesliing —_Atgontthen of | Hapfiela Nusa. —— a. — Step 0? Enihiatize wou to alia patterns he. uote obtained. fren. - fratning algo. using Hebb rule. = When the activations of the mets are ot. Loney then _pesporm Step 28 a & Resor Steps BT ae “input vector —_———— (SS = 7 Make the initial aia squat to thee: faput _veclér Xs 7 os u p= xe Le= Iton) | _ steps Perform Steps 5- EL spon.eath_unit Ye Ciere unl = Upolated th random _arcles). Sep 5 = caleutate the set —linput of the meboork § i 7 ae di | Step 6? Apply. —the ackivations owe, the net | fp ects 5 ut ty ere 5 yas oF - a Oe pee : the threshold and is no) Slept Now fecotback the obtoried _ pulpit Ye to < : Thus, the activation vectors are Updated Finally , test the network , jor_convergenee. Scanned with CamScanner_ Dates = QW Train_a heteroassoccabine menaory,. network using Hebb rile fo __sstore_input row vectors. C$12$2553554).to_outpuk __ pow vector t= (ti, tr) vector pains are _gusm as apt targets $1 dst Land _ aan eat won (asel! Firsk_inpuk woloy (2, Ay Hy, My) 240 to) (4, Y= UO boy (nes) = Wn (old) + %, 4 | We) (new) = wa (old) + aay = O + OXI Oo | wo31 (new) = wa fold) + wy, = O + Ixl = _ eu) (new) = wy, Cold) + yy, = 0 + OXL=0 W212 Linus) = We jold) + uy. = 0 + 1KO=0 022 ( new) = Woo (old) + xy, = O + O¥O =0 W032 (new) = Wi fold) try, 2 O + 1x0 =0 wy (new) = Wyr(old) +%,y, = 0 + OXO =0 Case 2: For Secwnd input vector (2, % % 94) = fb Oo 0 3) (2 eC ep wy tne) 2 sy fold) + ays b+ IKE = 2 Wy = © = o+0 log (mew) = Wa, fold + Yay, = Q)+©Xx| you) (mere) = vou) (Ol) € ayy, 2 QA IRIE I Scanned with CamScanner—— é = THe —Centia) Wiafoid) + ayy = o tly O- ine) =_Wwaa(sfd) + ry. = ofoxo tee (mew) = waga(ctd) 4 as¥g._= O+0 x 0 242 (me) = woy2 (od) ayy, 2 Ot! x 0 20 Neck, —4231_( ew) WOHL (new) WI2 (mes) = Wi2 (old) +2 W22 (nus) = War(old) + a y. wae. (musa) = (udga fold) + 25 Yn Loy2 (mew) = _Wy2 Cold) xy Case 43 Fowth input veer —_ (2% 43 %y)_ = (6__0. iki): {$,—yo) = {0 1) Gee won (new) = vou fold) + yy W321 (mew) = wailold) + xy, W031 (new) = 31 fold) + sy, — Wai (new) = Wy cole) fF tyy, Wid. (mas) = Wyo told) + yy. Woz (ew) tag (emo) Wyo (ned) = Final wl are t= wy 22, We 22 7 0g) Aly Scanned with CamScanner“QU check the auloassowabiie metwork fer dsput vectar | Th | U7. Form. the weight vector sith. no fees | Conruclion » Tedt whether the net ts abt. to recognize | usithy one msi —- Sol? Enput of qe ees = fou) =) =t 08 Testing the mnetwsork silly One naukecng _eritary. ” 2! =I Scanned with CamScanneree a — apy activation function we ree ef =u — uss Construct & test a BAM network to aocate letters F and F cae ssumple_bipatar _uiput - output _ ——veelOrs + The ta tr por Et (1,1) ancl fom ee oy _matn'x_si 5x3. The é : ige Gs 5 oe a * i aia ° oe * Scanned with CamScanner_| iW ait ae a doda | Sp Og ao 8 Slo bo Hs clodado Nt v wa | ul bea) “ERP OP Taya 4 = | SPRATT TE EP i a | fa + a lal : r ew 7 fe 7 i a $ a 1 x 2 ey | ; + flay |) | | a & | ry} if a adalat alt inl lana llasortanatlanslone ili ttc Ld Scanned with CamScannerpn abplyang neta “penton —on applying. _acliealien we gel 4 | _hents Presponse 6 obtained pt espa F depp pe ep [= fe ia Fae so Correet —apigngy atic gL a_i Tae : ___szeaponbe ib _obtaured. a t t =f} tJJoo00229-2.-2.0-¢ i : g2200 2 0 O22 fon 2pr-b-2 2222 -E sd 222) 5 wg atbation gundiona., ht get offi si acl i tty ct att Sptral Scanned with CamScanner=H Cerra ee eee a ee _Thus_B A: M__niekutork is _tonstructed and. tested 1h both the _clirections ____ = ecend Components. Of sto ssector. a? Input vector a=b4 tt 1] weight matrix ba_gusen by prey, i Cry rey Ee, M1 w=: 5s’ Cp) tip) Scanned with CamScannerL Date Binary, sepoccuentatiine por given input sector 6 [4 110] Wet Out Asynchronous —upclation of wodights. hare, tet —__tk_be“Y), Y4, ¥5,¥9 —— = Test input _vecter _usith 8 matsing entities 2 — ee -_Stepl i Tput vector i x= [0 0 4 OJ __ step a y= To 6. SOF. sol ~__Step3 2 Cheese unit ils acktyatens 4 eee Sob = Ys lH =O+foo 101 fe = pts t Applying, adientin ais 70 2+ yy sk —— ee eee Broadeasliy yy —te_all_otine unite, fie pat t oS --No_stnseng ka coc gee © Steps cheese nae Yn pr-updaiig ts acisotions til = Ry t fe gh. ees x Eto £ J oerit 0 £0} Scanned with CamScannerChoose unt a va gine = +B yruopa | — a =otfi0 1 oj | 4 =O+2=2. EE Yui 2%0 f2 He =| Henee y= [i 4 1 0] Lieratten, 2. o UT sy Stepo é ts : Ea aes — Steels weights ave initials ed IW) : 4 8 =} _ Shel —Stepl 2’ ow= Ti 107 a Stepas = Ell AiG 7 Seer = a ire a Scanned with CamScannerStep {cheese unit ¥3_for updating i ——parsatd Yj = 04 yun 20 * rik yy fon epdatiom ‘Now yes aa _Stept choose Bee - eee ce Siege ees = 2-3 yu Kot erat Thasee—y Sct pao —Stepsi 4 Lunt Latin z hoose 4 poe ap al YuAs = Ant 2 Yr; —Yyurd Fae e EO+ 03 “4 07 La 0 T ig Scanned with CamScannerrahén and gunctionad _apmeximahin neusal —mju2d_ceweloped + —t__Network wer the mast common nenlineatites such as Aigmerdal_and gaussian Kemet functions They are_atso_used_o es networks \ —b___Gauistan Function __clepined as a _\ Teaining Algorithm > = _ tep Os Set the weights te .amall_yandem vals Step Lt Pea} hin eo en eee Step 2% Pesfeam Steps 3“]___yex each input - : fap uni (20 pon att ton a : ignals_4_tranansitt_shh_the meat_hidclen lanes whit ee —Andiol basis .punchain StepB i Select the centers p04 Aachial bas pune Centers ane selected prom the set of _imput vectors Lt -—-Should._be__noted that __a__sajfirient_no of Centers. _.. have to be select ed to ensure set sSanipling of, Input vector space. aa Spiral Scanned with CamScannerwhare ay Center of RBE Fanik par input _veninbies or’ the _ysiolth op ith RBE unit , xf the sth variable input pattern. Step#i- Calculate the outpuk ef the meurel network » Ke got = Wr Vila e+ Ld. i=1 ea MOE of hidden lanyer nodes = output olut of mth mode 1h _putbut. Lawes er oii, incoy ottisn vot bio ita RBF unik an’ at nth oukbub node — Wirn — d_mth_output node wo biasing, ese Scanned with CamScanner