MALHINE EARNING Meching Leoeanin : rsa nen mpg gl a i poms. doo tarte by itealf ule owlernas prog ar | $e eo i wlo A | peag vonnnsi ny to-d10.$0 a akin ee eal i euisting dasa bas “Paine of_evaninitys thar é i up ordey sda pense oc naw“ capsrmelioie Components of Learnt) Prous - - Cy Data“ Storage i Data 1 —ssistsdd—un york dish CQ) Abstrarion -—Proiers extracting bnowledy “ tored_d. @) io oAion * i . re Types of ML Ci) Super - Models Ou trained using vabstled Aare wee dota _ with gi pur_to_yind_apchttai, ur ciion— pes Foy Re ime Y Clayy fication A fredXrs continvovs. Predicts 4 discntte clors dakel quadity 7 Prob St multiple > feeb wl Qor mor Clapsey Norinfes is a mul{t- oe cabled mbt? ~ clon 7 Nensungrs doipi cation) ) ) ; : Oty iepae_data uy_providsc pt ata ean taimeth unig —bnla bi ti heals ) + Vist te. Widdan patterns in input data ‘Two Lapse Ai) Clustelny Gi) Accociation }__Suhforiemint learning * : + Desired “Ahan inne ois awarded ‘ i howtovr oti _pumiched ; "fy Gaming Applicetion Troding 4A _finan us mp9} my ES ba it apron g gaaity | * Arobigs hiddin patterns Ci) Yate Colter con Ci) Conplsity (Cv) Date bia Gi) fon. quolity dase (iN) rainhnance \ Sal Dara Collect fae CY) Choose and praspant a training dota Ut Sy Data a Gy Sehr Age to “ites \ ) GQ) Train go to build modal GQ) Tor BG mods = (5S) Tap none modus in ial I R : Deretion of ML __ prows & Csreps 'N MUP; ow | ) ! (yD Modu” trasniryRon m fg Q aries ae mee Se dean yep pepeaumance ERVRMEDT — OF 8 —DE AAD — $$ Gy Long usivn Mastin Take wi, A peramire > detyel and predic ied ———— 8S \ o 7 Acted o gel: Fedivcta 2] | Now, 1) Co 0) indicat tovyeet owtpur - For modid dio be a vm Dre choy dbs high Actorauy = Sun d Totok Atul Sees NS No= \Q- Predicted YES = Yes = 1S No = 10 Now) fleoregy = tg $s ay siete —____ \a4g4 341 as. Now 5 var? ey.Grs peo Mure nylay Ueni}ied ab TVR sitive — some. woronyty classe (s'eok a ~V¥e ieFrue_Posi tive va ei i ‘Je Faby Positive Netty LO ot alee Nejosine C8) _clansii : ceuratyy._Detzal_how spin tad v8e_tipsp— NoveP Correct predicrrons “Total Prediction = TP+ TN DD ‘h +FP+PN +TN tt) Classification Deport 1 > Wa ap ort woncibte op Precivion eco, F | d (A) fercis in fatin 9p TP to teat te \o_predict ions = TPL xP Coep 4) \ P+ EP CO) Recalt Nencitivity Mit Rate” Ratio to, itime (ncta in ‘nel > TP ee (ocerél) TP+ EN c ici i. Roskiow- iva jnCFon = TN aus EF (A) FA sere + Hox ponic weer Precision anck Re cots “Wels Md ave p,L prow, 4 As orniny Of a_punetion which maps Tr_inputs te wp bayid_on ‘veining erates zg) = padi pasyor Seput applied “Machi i ‘ ped clay detailed amps. (C) dereymine Aatouer CK) Collect Lady “ine dasanas Gi) pit Thinclata vt (Wo _bruindng and tem {60 dali deakune 9p train {V\\ Apply tradniey dota ibd algo (vir 14 Dat Acct i)_until seve fs , epi @) Classiliconton 7 + Output dare ui tXuorm 44 coteyeria NM cm 2 4 * Alan's drue . (apne | aon - a pose ofa atyo uy mp int __wi Ascnste pare a Maly © te | ‘ 7 Fema be = ax _ a aie >Some Clatai_ oAgns WAL (i) Detision Tree Lil) suppare ecto Me caiist ey) Avi ott —Newor ot pwote W) Naive Rays : | 7} fins) mual a5 yam 1 _Hown , (é Classi} nu ss Spam | _H Ww)" ht} types of pails (i) oe ual MO lor umen — Imagen. E Reurescon [Estimation C ConTINUOUS Sense) "Thy Modth uh ured fo Solug suprertion pro dum 1 | 1 Jig ye eh 1 eiman duletinn beter dependinr variable Cy) pt clear ven de ble Cn) ng tine yf pate Pe" Ling Uinsay (Non-Une Aorre: + Randers ‘For © MW varioty * lostuh' beuitinn Tree. Atte’ bubs Byration : : fonawy ‘1 pat dune vari Popul Ayo = tim | |ry Yat blo Asta pow and Worm Cells whaler mot A wt VI i i ' Z elec Vay The_mai Iacrov wee Warne Wanna pe cr a pnt o1 which ee ee omit one dato pain tla pu highly corr eléled — w [exch Otnsr = - Overitting Algo wrrbs Well Ww] train data burnot fest 7 > Coosa) : elo ete Z 5 i Lo Fpivieras (eam > 19 20 20 Yo Se Raglan + Used do Solr Clarei}ication aypor ea tn Si och oferta \ ee) \ee™ SSE See hoecig sss es Ree SVM a"a Per tive mse: \ Sy, - ay meat am emst — f 3 ty +h) Ls MAE = 1 ely = Yl n mA UNSUPERVIS ED + Find pattery in uaadelid _tnpwr chal int aiounp win dede_and grasp Sa fois ; O Ciuatuing (4) Visvatirasion | Ce) bso Stan ba Ge) Dimunaionalty ddl ion — ; . ChuATar Zila gd d0TO gyn pads orn hai, of simi lor Kes _yuth on Shope color etc. + We neon groupe an dora -viele tie ¢ 7 Gi) Denity, Bared— : i i nos Cuties - —_———— + Moe high duowi ans _grouped i Np edn ype Chur Ws forme —— & & __ (at) _tnie raxrchice —— _—___—$ as “Dotu aar_divided into clustirs fo cncole tres Ure 7 Called DENDO. RAM, ~3 | € - “ Le Mean . Meet commenty uted _andSiupfe _______——¢ na . 1 Specify nou clustias Kut : : toa k dato prits” : * Comput tontreld- Stops ' CY}+ Setar no-«} _¢ Iu. ; 7 ay a, 31 ¢ idaon aii _@ra adiign to tlle ____+¢ cy): Find ww Univ old jtading ON 9 : 4 Er enets a nal de ato paitt bord on te ___, sag. ua hw wntvold+iS | ewpectation - MAX pmaTton) | | Useck_e_find vn enown vat ; th iny_ Aad variahhs >LATENT VAR... a. bld — yore obser vec a | ynsten_—_ MAU bbb hy alg s_ginds Cestindan man ui beli hoo af dotint vor ; Explanation » YS gy initial saan _ace_comidt ced | A ser of invonpAut_data en — (2) Enpttabion Step CE Sétep— bs data, Wwe timate or opined vahus wgproa'sciy dade, (_Masimietina Slop + Use —Complate data to > _porarnebrs : ae ae nn Conmrzity to solution => stop oT wise _uaprat utp 292 with initial ol ata -——§ 5) : ° arama Imation 07 idan Maxwoy : ioeasge parame estimate ofLL Decision ‘Tree learn ‘ng + Mh on appro mating o _ discret Values avg ep _cftun¢ Pre drey clans ov van 2} ter per varia, bbe Ts Supe x ‘sech_Leerniiy la Dana —— - Performance = Motyivor CGD Him) © = Se Ply) oy, PO + Moore of Aandonamsss f- Ly} 0. at tt iaehe ag Taw): €)- ECT, Y) T= torrent Safe - or ra _(s,A) = _£ Cs) = t ~€Cs,) Chavackevistic—_| Sprints tay | Atty ____} S60 Algo Crirerig, Type MV Dd. Tip Goin ae aa ee “ in _ 2 CART [tint Tnaun capa & vane palsy valve Auyney ie v ols, CaS I on Rots! 0 ! Yond bof (atsror.\ st Hold. use ey val homeric vada hse = _T or dint ud et, Iuie_sfor- “whch rec ts ma Ga) wane n deucaie area_node why ore wl man TG ee ec a uy Some tor, rent nolo as. bgp nosh ae los _unsmainins 9 ve_{i)_we yon OU pete ex _dauision _tver_hos old wWenJ_nockes Ee a Limitation s— eet gu enon —_anoptinsh—se} © Hardesty use _onemtinuove dota | Tndueria _Bfos 1 + Suto) Assumption That earner unes -b ouputs —af-glinen inputs Maat it hoo wot en couniey eh d : a polity eet dari Sion ree grceatines fem Eten de cl Waren inctanced assumptions ov Pmpee rakes about -yndar CX + tat, das RELY Ty Pepegee paccite Pte pis and odayrtadtea * Loser Approwimasion _Tedvchion Bias % ke , e pees oer ap a eran pS TSEC. Avoiding Over = Pern’ ara (a) Podv door Cttoy = proniiy eg admed oA dy (a) a a tonrenvovs__V &) aobility Ds handte ym sir dice, Wy 4 i wl dulferi costs + amet — ator dil ptiChri Coates gun) = \= (alg)_- Glsy = ove Gini uston'e, ee a ea = Giini-Lout look fae? = OP int Baden Ce (ous fe) = £ (v4) 45 lowe) + (0) Ww > 0342 For Temy ‘in Guna) = = Qiyy = Gls)" = os si (Tad) = 1 = ML8)* = Ql)” = ~_o-uus gals wns 1 - @lyy Uy)” 0-375 Gini emp) = C0" 5s) 4M to: 32S)4 6 (o-uys) = 0° 484 MY \M Iv Gr Ha dity igh) = GS-19 — = O:4e9 = \- ¢ A Givi bunt) = = (7iy) 0-Me9 4 (fy) O-auy ~__O: 474 free Gin (U, Weed = 1 - Gd arg = ne Gin lw Sttoy) = \= see 205 Gini (lind) = & (937 (os) =o:4r8Winnnr would de featvre vol Uo _ Gini inden __€ Duion Chosin sub nodes _w / ___ —Stevilsety —sce bull vst Of trea by Che ——awest_—gini owt a ae at | 2 pnis_dara vet criyst we Cale Wobadentvop y. iz ce 5 = O° JUD Ex = Tuto 5 = 5 dog 5 —= ; canes ened y ie cate gain van fat_enth 208 Sind your gos totty 0-0U9 > £0 sfx Aust ent Ot 9 decision for wege wind and for strong ; spp Cond Eley Ty Coinforlp= _0-0u4 O: Ops = 004gRuinforemet loo nivy Wype ef Mt din wren on pent (Hardware (Soltwmat,) Einkarauts Ww] _envirentmend crac earns ____— S involved __Q) Tahe Action — —_ (2) Um yrebace (Reward | Penalty.) Renin tn sew state (change ste, —_____— Navin bts “Any chorasferictic, no or pualily Mar can ke mvapurcel + is ps! _courtled ,¥ ‘hat com 't Jy} values ty ay UyNumerical_ + dinesbe & Coninvny: jeer ( (Hi gues i See ul ti J. A. AQ. 0 i ed Gatton r Numuicel 1 Nor whi meowremetor 10: hoe anvmviray huaning Rbuwh Vel Mat coh 6 © rovaled a4 % Whole vee (¢fud) calidon Aisrely varia hg lontinuovs £ Vat_tontain vane ep Uf 1 3.76 came eee Cdag oricol — SSS Volise —9f rico —voriabht une CeleTed - _ Yeon gt oup Of code gories =< bia = = — reat =_ DISeRETE © Wen he —0n 2 —— SSS Ovdinert ° can bk meonity f uly ov ole perk + Cake por ts“ Th! convisted i ingo no+ It had Pum, jee bale Nopinal as indices an ord Sanya ordina _ bat ne eS .rr—“—i + No Moahsmasical Vetus a i ond _cotsyorien (labels) Vox which contain nvnbys Ey fetuaty §5 Faika $0 MEASUREMENT fry ips. & object_= Person Yorighnllyes Tha ttt tp = S410 | Ro a alah Sass teatysh_of don btsined —___hy odupynnerr Scat *Caise) devek 6) entasunrna on . Oy snips with. variables ov. alana! ScBLE Vabyp ore mrigned to calsgorisees—or— groups wi] ___ no tnhtrent Order or Panbiry (ibe color _ Gi)_ordinos scolo Ni Osu to cat “with a fa lunes one ge pa. ies _br_g rs My. or Ordor bur he fined DAL OLA INT oe dif Ww Pata St me ee he 9p eg dy a «amy Vals BAL ausigned to oe ew oY gy ¢ wi ei valine ——— Bux no ¢solult_zaxo point such as Temp or time (oy fle bur not juatih y Nin} Ponto Scalp Nalun ox _errionec Yo catpories arproyt w/ ined wonsienesr—and di} blu Ave 2ero_point Jorwar! alee Enanaple, chonapde Height Dranyy seas Hh by McwrdDita Analy cig a Ci) peserip live Anoky £5 ibi ana Summ api2ing data 2) Diapnastre An ake iS = Sraty dog to leSermine couse 6) MALLE GrEnploraloy : yxy rok Cs) Mediviw G) Frese iptive (7) Trent = =e ; v)_SPSS t+ sear i * Eayy ty Use but erpensding £ Used foe tans i k¢ __vugrer/on ancly. is, Decision tree anotysrc (AL _R Free and opan sora Siariisicad analyse data ining pp tert ——Clurbortiny, Tews sniniay + Gon be evlindod wl pach ay, (3) Engl. Sparadabent too/ dor bani statistical analyst p inte, uisMalirotinn me wy A ay “Tey Afra — fi8 yre en eta Used for prog com Aston ps Raa wo sh pt Pandas, _ : re ety used efor My, gam dj 7 DmaTeRs = Youd yor —scisntiy ic comp ot _ Tasha svcheg Matric Lab: “yee pre aay > “inne ge way — = y_f! 1 Concbe util ined Yorba a. Vesgalinadion task whe i Shy + (6) enue tater Aad. __intejyaion Uihe dota Ubansing Vonstana ny moa cleangformediv’n Mw aly ete sb Open = cour, (an be wid. dola fi'sd. Libe tale. snaycin ML, visvali of on cess VALIDATION + tonpare dif ML nultaods een Doran “uiialivicled ive 2 Or nuns pov de Which 02) und te_troin and fest payJarm any > bop Cys - Veedation at Cero volidarcon set Jy ty pO Non af coda yet Tost 1's not _unta = pe dy clbbny Ard ute few cduots psfformane *ekyures jrdsl Jy fot bey jiftiingCommon form oh Lrons VotiOs fon () Tole Oy : Gy Dota Aivided in K epuas parks (pa gay Medsl trained on Ce-Wdo tei ernahdly (3) yo Tad _K tvs, w{_eath old tutd ay text (org ony —_____W try i do ARTIF Nevear Net AAn rowiriny onils are my To attonplish atark in BNN jr re cold Synop Gi Prowsing. Ut > ANI cmaists of nodes which as {tas connetled— Jo Cath othe t+ : ak Cy)_Dperisivo. BN ‘Fath, oth put Nos veeciews Crput s/9n hen sei g va op it Spa talultad Acad ion — eune Ff pitoced babi on welphted dui of inputs . 0 ; ivan : @) Wet gr Updo i Cach “dink oy suwrad nexwork Wu UpdaTad till ar yer ce u_obtoined (vi) Activation Fur C+ : : sail i yum 94 input 0d Git) Tp" vayent atid dutta for NN ____ ,—blw atilpR ol p layers. _ dations take platorin Addon bayer Ci Aenivi int “Mesrons: rr conn cel_nodss_in ANI : 7 ee Le —@ii)_nia (! nternedivte layer _____ AU .tomp uta slim ax) Olp \oyer : focuses ol p from ger up - Comparison _plw On 8 Conspyter Net pp 4 BNA CN (0) Spesd Usnuncing)}Sloulnitiziond) 10> (Fast _(Nesatecd 10°” (2) Sof Fc (ul 1 Cfaratles) Mordion Coop iat) (3) Sinek Loy wily Lary Bo: oy Newer (i9") Smal Nor 6) Neurons bp? Jusryetorplonity (10) { (anal ni (4) Stor site 1 onnett ian igarilieuieag ah dactote oO avcten (Sigurd mained) nt The de No Centra Contre) Unit Se_ Gatco (ontrel uni Ws dane ings sn Untne “oe oe __ PeRLePTRON Mobep A ny modu gp AWN Consists of Vp-vatas som_ahd—activatiirs suneion. bi 0 Ui Paramus ada + & — Bio Seen nauran «i au dung abwiry beffor TF aTG “by sont rondor vinitiad yolbaes and conipane tol actvat opin Araya in pin tongs daly i ea asp Ba pind gp folio ope to wsigts and adjtat them “Artovaitegly dit linn 2¢___ Dprsty. pa ; pi a oa BW viewin—d mr une “Newrad Network ee Sout ee Confro)s_ho wo puichly scams rer z= ze - . : sum _—__hyinition Catpur = - ye Sw, +8 |— giv i Signaid Fone 7 d is a Qu \te® Laan nop Aas SL sle teed aa tatad Sar o___ SWIM —Coupport Veutr_ Machines) m * Sopuvisod Learning Technique used sot Clarsiji'cos. ‘on = an prennion —ataainly for clesif cto (4 (s) *thain Obyecyivet Final Mypeplert fad dustiattly — dorsi fies Uf data poi 2 * Dafo_poitio- Worry to typer pons and inpsnts a position Nig. plant ah (MG esp suppert ventory?| a PM ad aif ora mat - tty par Plans Unseen ali vides dota info At | hantnr —— cloner Respir 2 +Gop blu 2 linen _on_closest_clesfa_ points yp Le ge tapgi ins enone to ts _bettey a nob magn EER a Sy pe Linear st berision boundary 2 Lintar £ Non Linea | Hon LUNEAR plan + Con dat ’ ee ray,