vie eS OU aN ae TUITIONS For Full Course www.Lastmomenttuitions.com| Module 1 1. What are the issues in Machine Learning? (Page 8-9) 2. Machine Learning Applications. (Page 11-13) 3. What are the Key tasks of Machine Learning? (Page 2-6) 4. What is Machine Learning? (Page 1) Explain how supervised learning Is different from unsupervised learning. (Page 7) 5. Explain steps in developing Machine Learning Applications. (Page 10) Module 2 1. Explain Regression line, Scatter Plot, Error in Prediction and Best Fitting Line. (Page 2-4) 2. Explain Concept Behind Linear Regression. (Page 5-7) 3. Short note on Logistic Regression. (Page 8-12) Module 3 1. Issues in Decision Tree. (Page 7) Module 4 1. What is SVM? (Page 1-3) How to compute the margin? (Page 5) 2. Key Terminology of SVM. (Page 3-5 // Tuning Parameters) 3. Quadratic Programming Solution for finding maximum margin separation in ‘SVM. (Page 6, then on Page 5 consider the Lagrangian Part) 4. What is SVM? Explain Hyperplane, Separating Hyperplane, Margin & Support Vectors with example. (Page 1-6) Module 5 1. Short note on Hidden Markov Model. (Page 11-17) 2. Explain with suitable example the advantages of Bayesian Approach over classical approaches to probability. (Page 5) 3. Classification using Back Propagation Algorithm. (Page 8-11) 4. Explain classification using Bayesian Belief Network with an example. (Page 6 -8) 5. Explain Bayes Theorem. (Page 5-6) Module 6 1. Explain in detail PCA for Dimension Reduction. (Page 3-10) 2. Describe the two methods for reducing dimensionality. (Page 1-2 //before subset selection part)a Module 7 1. _ the essential steps of K-means algorithm for clustering analysis. (Page 2. Short note on Hierarchical Clustering Algorithm. (Page 2) Module 8 1. Explain in detail Temporal Difference Learning. (Page 5) 2. What is Reinforcement Learning? Explain it with example. (Page 1) 3. Model Based Learning. (Page 3-4) 4. Explain Reinforcement Learning in detall along with various elements involved in forming the concept. Also define what is meant by partially observable state. (Page 1-3): Modwe - 4 Whol ¢s vockine Leming @ Machine leaming can be olelfned os a brah q Avsyretal Intelltgence that hos He abyliey to leaun on Fs own utthout hafrqg all Fnforrnadfon eth Hespect to « domatn in the Program Hse Moeline lawn % ao ty pe on depreal Inbellfgence (Al) Hat provides computes” wth the obfirty to lesin without beth bp leely Prog rammed . Machine leaurfng focuses on the development d Computer ‘Programs that can teach themselves Lo Fie anol thorge when exposéel to pew deda, Mthe Process 4 machtne leasing 9 s¥rflox to Lat dota nnfrfng, Both system Search Hough data to ook dow pattem « tones ins tead extracting dota for human Compre henéfon as 9% the case Sa dodn mini ppl cadfors , machine leasrrg uses tok data to impr the progam’ own under fondfng.. “Machine lean nrg GPrayroms detect otters in chta and adjust Prog vor on ocordinglyKey Prgnks * ML twins Dota eto Frformabfon Crtearfrgful Dato.) "ML 95 combination Computer Since , Engineering ond stadtsdfes j a "ML Ves Shale Heal concept Jo process olala and to Solve pre blerns. sw Tntewprels later ond acs on I. “ML Pregrams a Computer to oplimfze performance Cribeuta using past expeufence. Kk eH Tountnolo ies Kapet System 2 Yhe expeul System ane the Computex opp Peadfons developed fo solve Complex problems fn a puttoulas domat, at He devel éxtra- ord inowy human Prbellyence , expose and clecYsfon- makPrg abshity 4 o human expeut. Chovcrcheuts des | Expoet Systems ° Kish Pevlormmence # Undewstan deble ° Goltuble . ° Heghly Hespon siveCpourk System Ole Capable # Rates ' © Induct and assisting human 9n dectston mnoktiy. * Demons brea . © Daking co SdvdSon. ‘ Diagnostn ; * Expletnt : “sey Prout ° Predicding Hesulls ° stn the conclvston . ° Scigges dng akeinadive oplons to 0 problem “he Components 9) Expentl System Sncludes - * Kowled Base & Knowledge 'S Hequtyed to exHLAL” Thdeigence . “the svaers gf any ES majorly depends upon He collecdfon v ocewate and Precke Knowlede d “M * Foctual Knowledge ¢ Factual Knowledge ?s that: Knowledge a) the Louk dlomain Hat ts wutdely shaved typfeally found 9 tecbbook ov Journals , and common: Oprreed upon by Hose Knowledeeable fn the Pardfaulax° "Hawést7c Knowledge & Hewatste Krowlede 8 the a igor0us , Move experentfal , Move Judgnrental knoulle {ge 4 poyformance - In Cordvast to qocteal Knouled gé , heuststc Knouledye 5 Howely cfewsed , and 75 lugely Thatvidualts tf. * Knowledge ueprensentat-ion 2 Knowledge Hepreserdatfor is the “mellod 40 organfxe and formaltze the Krowleage Sm he Knowledge bye . It 18 the form of IF-THEN-ELSE dues. cihe VE ple of the ule ts atislted ; consequerdly. He ten oot con be Concluded . PKroulledys Requetiion s “the success yl ony Cxpent System Maforly Clepencls on the aualely Completeness, and acewney the nformalfon stored $n the Krowledge hase. ‘the Kriouledge bose %5 formed by Seading from venous expert, Scholes , dnd the Kwoledge Engineeus, 2 hitexface frotne % essentfal $n declucting a Correct , firuless solution, Lt ccaitres and rnanfpulates the Arroutety en Know ledge lose fo woke ot o pout ewlase oludfon. ¢ User_|nfeu provides Fnteuacton between user the FS Hd the £5 Tso. Ut expléin how the hos apsfved ab o pul lew Hecommendaffon .Si 5 Sek 2 A sek of Cxomple sed or leaunfng wb i the porounetens 9 the Classffe. the bretting sek con be Elected by opplfing & Handom fee do He dota eg- Seat 20% 4 the points 03 Handor to generate the Tmdel ond “tat agetrst the semoiring 80% Validation Sek 3 R- set exomple wed b tune He posourn ebexs | O chsglée1. His ysually vsed 40 adjust the Chssifitakion Porornelens Gn orden to avoid oveyftirg. est Sek 28 test set Ys the dodo whose outcome fs Alsieady Known and i used to deteumtne the oceuta the machine ns algor? thm, based om the eres red “Tosigel Vawfible 2 [k 7s the oubpuk to be precicked pom & meeliine — lewrfh algovtthm. IE could be binosy Ow 4 You owe Chessipiirg ov 9% could be a Cordfrvow vartob fe q You axe deing c Heguession . Class?ReakFon ¢ Classflé catfon fs the Problem Yen bry fo woh&ch q a set 9| Categorfes a New Obsetved on blag , on the basis of a acing sel. of detox ContoSining obsenvedfon = whose Code gory memboishte 43 Known.‘Keguessin 8 esston onalysts fs oO stodfetP cal Process for Loe the Melabfon shtp among Vaséiables . Feadwies_ s Q -fecctume 95 an Individual meoswiable Preposty 4 & Phenomenon being observed . Feotwtes Gnd obbtbutes csie Indivfdual measurements , when Combined with othew fentunes maker braking Counple .‘Types of uchine _lewsiry 8 Lupernised lewiriry & Leavin? proces uith the help a G teachey %5 Called Supenfaed Lewirfiny. In Supervised Iesiring the alla ovfthrn geneates a furchfon bret: ed Fopud: ty deshed outputs. Ohe stondaud fonmnul odfon 9 the supemisecl leosintng task 9% the Classffication problem : the lewinen Fs M@ulived to lean c Junctfon which mops a veclor nto one g veal Classes by look fag ot sevenol §npb- oveepud- Sample 7d the function . Unsupenised mina Keawrtng Preeess uukthout ‘eachor %5 Called UNSupansed — locyarre Doty eee the machtre eau aloovtdtn Hecetves the Snpuds podkeun and owjanixes thee pettons to form Clustews. When o new ‘input patbern ‘S applied, Ye ofves an oudpet Sepore Fncltecrlfhg the Chyter to whith tHe Srp Pattern belongs. in| owerert leaurfiy 8 Sinton to Supowvised lean’ be this leans Process only CaklAe Fpformadin 4s Gweflable nok the exact Spormedfor. the leas Fr Process #5 done based py the caf lf. ftyormadton anol a feedback signal called Ey orcemert Styral $8 Send boek rem outpud. to the Freud.| Losues Sr Macltine_Leewwtfng_ 4 Whed: alqartilms ertist der leasing general Longet {20m Spee lie trafifing example & 2 ty what seltrgs will potieuler alo wits converge 4} 9° 9. sufectent tro date. 2 ne desred function , ge ue ome 3 Wh%eh algorithms Pesforrn best Jor whfch types a Problems and Hepresertadforrs 2 Ae How rach befiring deter $s sulfetent 2 5 Whee General bounds can he found to welede the Coftlerce 99 keyined hypotheses to the omount 9 txainfng Spederce ond the chogecten a the — leaxneu’s hy pothests Space 2 6 When ond how can prior knowledge hele by the leone guide the process 4 generalictg fem example @ 4 Can prtor Kiouledge be help even when FE Fs only opproxtmnetely Corveck 2 B® What 9s the best stredegy ov choosing a usdul Neck = fyok expadence , ond how dow the Chote Uis stro y alter the complexé be leaxnd Problem of Od a% Whe % the best wou fo veduce the —lecwiring Fesk fy one or move qenckion approxtmatfon problem ? fo. thal Spechfe typfally oso buy yy and Y the fs o customer who buys X ond does not buy y he or she Ys a potential Y Customer Once such tuslomneus ove Fderdslted i they Can be Fauigeked dor Cross. sell Bi Claer a. * Classi iteegon_¢ Clussfzeatton 9S te problem Fdentygirg 4 wl qo set y Colegote a new obseuvatfon to which a sot 9 Codegores “a nel obstvaitfon belong , on the bast a trafnang “et q deka conkofiting obsevation whose Cedegovy nem kevshts fs Krow F Pattenn Reogrtion 3 ts a brarch 4 machine lecviring that Jocuse on the vegn “recog bfon 4 Stimilaut’es and vegulantifes in data, Optfeal chamocker stecoqntfon $ fs the Hecogni tfon chavackert Codes qo thetr imoges. In 4h¥S care thete one mulbfpke Clsies, SF Mom Gs there ore chotacdes thet % Lo Hea, Heoogntzed . Hed wf Hecoy nttfon g Yeaple Style | choxaters may) be wrrtlen small or laxge » slanted vstth pen or perl, ond thee oe mony posstb] imog’s epee to the same Chonacker . howe ayer ert handwatt nr 'yfae _Mecognitfon 8 the Fnpub san Fmage ond ‘the Classes que people to be Hecogriized . the leaurfing Program Sho Jeevin fo associate the face Images ty Fdentffees . Medical clfagoosts 3 the pls ce the Helevand: Shloradior Aout the Palters ond the Classes ome the $llnesses. Speech Metagn fon & “the ‘Tnpus 5 ocowtf and the classes Ce words Hot con be utteved . “the cssotiad?on 4o be leaned fs yom an aCoust$e stgnal to a word fi 60me Tonguages . Ne Nodusaul oressing §%s a feet 4 Computex Stance , otf} cal ‘nkelbEg ence , ond Computodfonal Tinguistes Concexned yutth the Frevoctfors between Computers cand human Cnabupal) languages. 5 Bromebsit's 3 Ps recognition or cuthenttatfons Using thetr phystolo: Rol andlor -Hite— © “tl explatre — ite doin oct whfch—en—explonatfen -abad: the—protess behowfostal chowacterkstes thet tequtres ayn Tdeqedin gf Tapule fom difjowne radadtdes 6° Knowledge Extortion s Leasing a stule fiom dota Fs Known od Knowledge extroelfon « “the Mule fs o stimple model thot exploins the dota, Using whith on explnatfn about the process vundodfg the dota %s genevated. @ Feple4. Compessfon ¢ By fitting sule to He data an explanadfon that 9s sfmplex than the dato 7 Sequtaing less memory ty store and les Computation to precess Can be generated. “his process °s Known as Compressfon. % Qublten Detection 8 Cubltex clelectfon % the process finding the stances Hal do not obey Ye Jule ond aye exceptfons to the Standard Cose . 4 Kesresston, § ° ‘nee pession analysts Ys a stodtstPanl precess for estimeding the selatfonships cuneng volttahles . l0- Density Estimation 2 ‘hee % a stuelwe to the MNput space svth that ceukin patlerms accu ynove gen than Othe, and hosed on the remuency y Aceunance It has 4s be analysed that what happens Qnd what does not. One type ¢ density estimatfon Ps Closterding . “Keal World pplicadfons 4 Machine Leow | ¢ Web seasich . | + Computedional Bobgy + Fenance + £- Commence6 Spice Lxplovadion . ‘obwolfece. * Tnformedion Lelsaelfon. * “ext ~ based — senlfment analysts in soctal ‘Netuawr ki * Network Frdyuson deleclfoa ond feud detection. ° Emotl spam easing . * Pedretion g eaysipmend: —JoStlusies . Hing uit swle —creodfon .Module - 2 EEO wilh Regression Machine Learning Notesfe NeSSION an {std “Req ression analyst 9s a stabfsdfeal techn®que , Lough — voh®ch ~ the est¥madion a unknown Vewkable gem the Known vosttobles con be cone Uhe Known voudables whfch fs used to er tfynade the unknown vesfables ts Callled rdependent vertoble OF Bplanalory vuutoble. ‘the vnkrown varlabe or the variable Joy which the volue fs f be predided is Called Dependent vartable ov explained ‘vouable . ‘the "ray vessfon analy maybe _of the following types d te Simple ond Multiple 3 dn o simple wegrression anolysts the Numbex 4 volfobles ‘Trvolved ull be two} one dependent youifable and another Independent vostable , On the other hand fn & multiple vasrressien analy ss there will be more than two vuables > one dependence vurable and other ‘in dependent voufable . 2 [neon ond “Non neoxs When Phe dependent vascable moves In a ftxed proportfon y the writ movement of Fndependent yqufable St %s Called |erea Heqyessfon . A om Yon- neo or Corvy- [Freak Hegvessten the value dependent youPable does nol charge! by a Constant aAhaeluke amount jor vutt Cronge tn the value 4 ‘Independent -yabalole. 3+ Pedal ond tal 8 Th a sonulfple regression onalysis 4 only two variables oe analysed ot a time , Keeping other veutoble os constort , 8 9 called poutfal Heqresston analy sts . Th oa total Hegressfon analy s¥s all the Snportont yorables ate constdered . Reqvesgon analysis though ovaghic method . Graphic method Jor vegression onolyiis may tate the following tary dorms. 4 Scodden dfagrom ov Free hond curve method. 2 Method q least Squanes. Free _hand curve method Under thes method the value gl potred observed the vdables axe Pbtled on a graph paper . Yhese points gre o pretuve q a scatter dfagsom wth Sevemal petrls spread over. ‘While pbtbing the ponds Yndependent vauoble %s taken on the yaks . Q Sagres Ine be duawn %r the between these paints by free hond $n such a way thad:#41 % equal - distanb all the pofrks on tthe side 9 the |The . * % ot the moxtmum posstble newer +0 oll the parts on the graph and © tt passes Hnsiough the — Centve 4 the stadteved —poFnds. as Fine % called fe Fhe of best eb fxample- CQ Megrresston ne Yon x. Hae x % ‘dependent ond So X 9s taken on xX axts) x 0 6 Ah BGs Yoo 20 12 2 58 lo 20 30 yo 50 60Wetted of least: _Sqywes R weqression ire may alto be haun on the hosts 4 tie principle least” sayanes . ‘Recor ding ly, the regression Ihe should be drown Hatough the plotled points Tn Such a voy that the gum the squares of the devfes¥ons q the oclual ‘veulues of dependent: ‘vaystoble prem Ther computed values % the mintnom or the least . = OY Ye)*= 0. Hee Y stands Jor He Known value @ the dependent vafable and Ye stands for the esters Volue 4 the dependent vasebable oo bto&ned from the Tfne. Q bre feted by thts rnethod 2s Called the [tne bet Jeb. Yt ts becawe , with Hg) eHence to these lines, the best estfmates te ‘values one voutoble Jor the SpecHeed values y He other vourtable Con be obbagned. le__/eneou Regyesson Model “the stelelfmship between a Dependent Vaufable Y ond on Independent ‘Yufoble X % postulated oy a | model. Y= Bot BX +E whe Bo ond By, one Constarts called the model vegressfon Corl Cers or Pauamete¥s »and E fs a Handom Astunbane or ever, the Fneay equotfon Provdes an accep table Opproxtmadgon to the true elobfon between Vand xX. | Tne owe