Scribd
Upload
177 views

Machine Learning Lecture Notes

Course Content: Unit –I Introduction to machine learning, Data Preprocessing, hypothesis function, machine learning models, supervised and unsupervised learning, correlation, overfitting, underfitting, Linear Regression and Logistic Regression. Unit –II activation functions like sigmoid, ReLU, etc., weights and bias, loss function, gradient descent, multilayer network, backpropagation, unstable gradient problem, batch normalization, dropout, L1 and L2 regularization, hyperparameters. Unit –III Deep Learning, the curse of dimensionality CNN, flattening, padding, stride, convolution, pooling, loss layer, Googlenet V1, Inception Network, One-shot learning (Siamese Network). Unit –IV RNN, LSTM, CNN vs RNN, Time series, Reinforcement Learning, SVM.

Uploaded by

Vidhi Sethiya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
177 views

Machine Learning Lecture Notes

Course Content: Unit –I Introduction to machine learning, Data Preprocessing, hypothesis function, machine learning models, supervised and unsupervised learning, correlation, overfitting, underfitting, Linear Regression and Logistic Regression. Unit –II activation functions like sigmoid, ReLU, etc., weights and bias, loss function, gradient descent, multilayer network, backpropagation, unstable gradient problem, batch normalization, dropout, L1 and L2 regularization, hyperparameters. Unit –III Deep Learning, the curse of dimensionality CNN, flattening, padding, stride, convolution, pooling, loss layer, Googlenet V1, Inception Network, One-shot learning (Siamese Network). Unit –IV RNN, LSTM, CNN vs RNN, Time series, Reinforcement Learning, SVM.

Uploaded by

Vidhi Sethiya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 54
| - du p dam is | | nla ream a ete ae 3) [3p data ag well a 1p dake F | ofp 3 Voyhed to | yovied fo mate! | modes - | | 4 Dead ts to Kaen te | ite | mL ; ea, to uF oa - pe ad | | podict aerate A ap ashen Yinerr foe wipe Abo data! pete Tent tae aug fade: a ; epee dosnt ved ee Ee tiainey foe CORRELATION Rornabelin cam be AncteLedes gpsabdes ate : pope tus yarbles . Ww one vrarinle epfoct % CX ae vattele thor the tw? | vaui : bach © be TI! corllasor . ae foo Aerelhin : Neyabive Correlabion (ae sn 2 a Bo Wo 1 |purleblee a ee urdagt eet ir when the Choe hs ely ha oven for tainting ete i el de este sop ta abyou' Das Mae — tHatntng > feo Wea ge Tooting > ook» Vavinuce —» Hegh Tus is called Mele beng : : Reduce volt, ee will ke too Hfeergic 4 | peti ie cuhen we tegt Ub Good foe: rand tion hen —- pam le eniiinibissiriate Unpletaoteet Naat apres: acces tlle predectsn « Wiaedivas 2 : benn » Lovr ION i Variance - Low-° a Seema 4), Heels Gs cet bos ee gprs o- rake Shout be sablee val PO teil tay yo wale LK . 4 UY bast, mithos ured to ow Mines Aebtunsbip betaseer ¥ andy ( tndependine & | dependent vawerd ) . g 1 Bu jt Une S found uel Hats eyes Ih sii ‘he Hoi aii lett yab Uae £ putuone polis amu th Wt rwintnar ysmXtC sD) Ye pathad Keb »\ Hep ae le es eke. _ Sept date. pupation J Deka pre Pacrortvg Ql Hing date into hadr_/| but I walihetion bet butling a yoolel hypotives : : gusctin es fuvclrin ee Y Gt fs uot to sack pobul minima by. pAradcet Aucms conveyer at Min” eres. dnd rardar Aethvalive + 0B RO ween : bo welinb’ el Wud with divoute fused mith Cowtiininy, > data veal dap. ec eee an A 2) welke it whicle oan | lL w on Uden »Cr034 Exner fre # ge = i a a MSE a low etic ———— fae mulhiple local ménimas. Lime We nooel globl nunca errand Loreont il! apes into gobak os Ve pr cow. Page LOGISTIC REGRESSION . a Db mod dog sale oy eae H Is wed at fs CTT mice orn pees daa bs | EET betMst OL Prample — \Emnadl Spam of mor 3) Tranetin draed or nt TAax = STebs wssating LR todo: ay Reading re das) Quod gue deter 2 Z = Who {We tt S) vali pet elf [ee ds vl betuth O44 - ee ( Dror | ~= tee oo) Cour fuer ~ Qt 1S the otp0r beteeer fale elt tel al I Oe Ehhoty. (wt duunetion is Ubed a tine) eS oe NN > Wan fowl. 2 == = (2) a ee ARAD| En) 1_ DESCENT Pat : LE IS the inet — dlabink © eyed reksice (ahs functions descout i Maed soweh tts Lobalurlnine. Lip movin srnalt seeps dascouding dourk + GP Se ) towatels wininun Steps: to find ent fpadiont drgceuct fund out partied. derivatives . 2) Chick for tek Ve valuos ~ Qf tre value is tre thon decuae praneter . YL tne v lure —Ke Hun inerzare Parovmeters . U ] Ss) The Deadu rade () thauld be kept reall an ode de avanh muna wae beep thr Learetiy Jute hugh Hoon ct vil Fe Langer np wl neh Fench Yaad yuna classmite pile es Gout incon! Capt hunctiin /exaot fe {othe meaacut % hawt onsen the mater is prodicteny . the peltion ketucew- dapat Ct) & occtpurly) Sigua Actination gucci cd iS ame sti fi aed : Ped woe = | Oe ey Set Lt ani Hic allen called Loyivic Luncion . Ltcosed naps | Abatul yale oA biluecm 0 aud 1, : den) = oC) x ot) —_ ee aa ised Lproer Lud - phe hk wMeo . ee 2 ae et cit» 4 lem ot foluw Oa Rol danni octivade abl nes at MASONITE oly Cotbed neurow ae aotivaed Relax) 2 ran (0, %) - | =a | 6 | disady >) not goto amtrold ‘i on re i ued Ow fase i lia Mase? Ul wobebhities =f We" paw Laer tron fro aceWracy OWA obek “Wilh Cows L orton corll be mote - te The prorkoms ast Apu dale ane \ pets dake — missing dota crneks gaps which tip es eudedor anys cai hapa in ate fie overab [3 AOE ro E oupes beg on a «+ datallis fiom Vaklod Sothots « Macy Liihs Pe ~ | wu ca | ee : “ee | *haduck tre amount oy date ~Aoduie os cornet Coe (i i or i uy) { ) | a fperaldne oa } —. faduicla sie Midi tt on: Ml ag a a ie 5) [ 6) — . = e oe | a 3 asa a 5 ‘ ot Map LOA eet ~ t Ata is dover ent Kae ! vada at otha end. 4 Pi wo | fo_nnulticleer ae ; pegs ‘ Visas lem complex Men combler ie & REGULERIZATION Repl ; 4, L Leen Lr I a tefind vecten distance] Manhattan Lik S Linom = Tal = [ule bl aay HC a [3 el Vprttien k= ANG oe TT Ly novm = = euctitewm dirt | die prom gvecer. V Iv] = Vee oes EZ BME Mimspuy: * fieagy EN d GX Ora dient Desert . i: w) To 03 = a (G - x; ae tN ee a alk & ; M= all no ee Mie Word ao uf um Sin. — ero i s ae ee NP tends we = ee ae Date : Page ‘ Stochastic Sradient Aereeut . = 4 Sample fs taken) es ans bk 1 Sample | 2: only J baring sample Is taken. thon punial derivative /'s taken . = || anpteek ep “pale tsomple we dyvile Mies ant ch up Taye thos batches , “| why, chr & noe ANN 2 doe age plea iaas, ee | The more tre Upue Hata } u ot LU ous eee ae e : a } | ENN Soles thls problen - Cnn does 1) parrot. “nent =e Capos act ORL glarsdl by tre ontine 1 sist pa eg de H SR Da ene camber apt tcad ee) Ailton. pppceliragss ~ Srolucse the Aimunaiono a Keun moc . 2) Aotcvalion a 3) footing a 4) dally, tarnnschedd Lager (HIN 5) duet (ayer ( cofeman) CNN is Uaed kon 2 a Image roooynstion a7eof V olaserg' © viyloe datictine face reeapn ili anput faym ~The input layer Iiguisie Be ita Uiv-the bon tf purleo. the dat In in fle foun ad matLr OONVoLUTION) Revie negs : Staltoh Local i ACTIVATION ay Ree in eovwideration eles Cusnetins Fnee CMe tlitier Is a Minsar eperadics ee tin Lineately we wae set, . | L) Ze tee ec ee amd replace U_ustth 2or0 » faving. ~ Wbecnpig | haurrsanling fooling fe done & sade the erliocted Lean | rorlily swdluew Sips | Cex om onl, Lingecttoaue elute Be kop quater a0 dominant testis Lip onrtogi thine = Maninuum value welt he teen tacks Uavk op deakist wap Aw mooldy ° Sea Ont Hoek é “nay wl be Faken + p the Vine (hb width) by haty - ue v he rutwilimanedrnad tabu bing b Cation | unto & Vingl Covdiows ‘Voor « 4H Thus ie fod into Leraicted and tr Ifo inde aly “The Ec lane is ubeal to Wlacat fg He feclitue Prediut by Conta 8 oelin ee fe mone Wy npuswn In-pitbae lager it OOnhiches [49 each try nett ef ryt Layer Ourpur Qin te fully Comndeted fridline OLeégut Levger Uap rer Wind the eulpuct-ao clnereo % japon OfP is ginem by, one het mending » a Slee READ he pine by rh wledeamenn Hin eA 91WA Hoe Lipa inne dele mone | Db, comrtlution, we Aplus d ther be YMA ee. gO (mage Wink“ BG In te eo age onr uaed Leo Cow C =. 7 i volo jee te ea _—..:. fading br Ame iy = yale gn Cie eee ye = oo Output Sine Panme ao ree 4? 6x6 - oe cae ofp t ol (w-f t)) a pass » ostninel ing sine ?6 7 Fy xb ue 2 8X8 — ax op ‘ae 2 be ee batt $$ — oite = pagal these iis t gets wade Le es inp ~ dtvide buigh din High Stride Corre | pitch Jo Cones ix (os - bd out > c . [ade "4 “eto x ‘tp ee — tis a Avtalr. tase pe F Network || akc > Cnn) arch .— || 4 va lo nM bt? EK Si Ne- filles ; 7 9D ISS io "seis one ha a a 2 Drouuafot Wearing 2oly oe I6 ~ [BBM panarnebees- 1b 13 CAIN: Bee 2014 ek - uception wudel : naw ' oro = Lewsey | x; ‘bes. X fee ws ae 184 pole ix | pevarohittion Is Used before epblidg Li lbens ubtamtott’ , slang nul cempine tim Wack —> iponallel bwers of frlfers Wet Poon 22M rulbiple Mmeeption Hecke 7 teil lncrtiens \xK\ 8 ae \ a Bers. ae the teoburlgue applied by ai] Is red reduce the | wink by OI cone a a used to seduce Cornuputaliing Dao ott Aaluste cose Sn «|| Rb ie woed in Qa tbat PP les Ba s* es Lwosptin Block WG voto au dnp ntlgstone ued by google Oi) $$ || ProRomp with, YOO. we Voky deep tun ane prone LO overpeticry Vs digit topam ginliont updati 4 entire weet Aiea ont batagea hae Lnggeamitern Atediibated ~ diary we Ue is detgpoult-te Selecta Kaunel yao ei wy by Heder Yoho Unf Is dirritated eae, Ie prefered webu fe js didbilled Aycally | Unboptisn. architecture pelweo this pLeblrm o> | Si allows multiple Wp psaild |ilkrs to tenes | Wi Hie aru La dap NN AR compuba Ronwolution udu die a ———— peewee | | eootaal : = Ea, NN Herr fs is toulemcy for : Su tct pict 2 haws pet updates bactuads . . ene layers LOU He i &% waded — Thre i$ vanleh Om de ww: Resor aah Lee to ALdohue eas wradiont : ae Seth Admins . p eormmuctimw the acaivetion Of perious I Bld ford Cage bo ee Roane Sktp Lommection : spatcad 8 TVIV oo i ) Rs e ce a x SMe. _ Aigeronct ) row ieee secre prt One Shoe oe classmate Me tated hugs dakaa ot foe duap Lavneng mae ©) but ‘ 3 we ‘ake ort iewage paet Alla «TMs jmabted ont Het Cogn’ || thes wader _prvolt cts by checking the Soratlewhg Y/ oe the | : [egg pane Wo ride CL HOLL Samachar pert claves ofp will we a number between o amd | mine 4 Bae cle D gorru sans. claro b 0 indtertes Hey ore fur Aff etna ohana = Tre arch ~is srnclar te Hot of CNN. You diphetimer iSHret rhe Ann layers are teplaced by Adfotener = Vt poleutster divtance / Ae Horie "4 ale A 2 wee os{ fun et Wu nae af Sanr hk Ue synod pes peeps CNW Batch Nowiraligalien —piiaaeadipetibn haw. ol ne Aforent We vi faces pee - a fg ts ; Vad cakes : I tse elie ee a meee l il TN a_ilag date to eommen_Sale writnowt dicttind Us Shape used fo torweyge dase footer. | | | | | | | | | | | | Tweaks wn pater. Ord stl addirg 3 nye Deg NN ; ot Yo 0 andardli reo novels ¢ he LA Oo) a J Vas hor % RNN ¥ | ere 40 ome = I x Bae pelncing => Music _gencelim i cc pis 3) | many t2 one ~~ gertinent. _, Weethet forecat. 4 Mons to wai —) “ase (bi Z \— i spain Wass WEL gett | } di 3 v x om an aehivelin Sj = oO CUxetWe-) te) or, I 4 =o ( Vc. +B) bel 4 Ted Same. es IL Wee? $ am eo i pi t | Bp EA @ e x, 2X] Cy 1 peers fel meen z. hart Loy ; Ly hitoricer Bala f¢ not Aeguired for prcdicten. only Uating seme dala we heen Ee 2 Saete 42 Sate Hees | erg eB safe Sul Coney Dd Oeetput ee be 4 | gtermecliobE coll Hate Problem 5. Col Stake Hent Yade Fg eyeegete 44 wl wey xg | t Wi x%) Moa Learn. i fee i: (wy x S,-b +r Xp) \ tr Poge —___—_ \” ah date | time vom parr - 3 AWN hoo Wels The Det ole gus Bo ae ese eennictes {ee EN erly ep Large ae | tw taper tn Rvs proved omeiy me {|} ene at cach time Sep” I I aa ey ay slg usr He gut pie Antilden cenit ‘a Dees Boe ae RNN ANN tld fo hame mowpty as i _not only ° nbernalion prurcous hager ae Ra epee eal “A i + a Mang to Manny - FAkio Legysme Y ds as eupput anes gine seyusne Y tinge #2 Culpa franplilinc oe Q@ | Cvw us RN CNN | Rwn’ Ap phixo eowwntu Lory RNN (eee time Step erie ee sunlit le PE tage | 0 V Wrote) blet: foe [eh rer Pepa plaosipicatan [time noes ake . parcar Hidden layers purl ouput & poy hithhon is not sullabed Sah 5 towugesud ot lage tt | desonts rear inte pn butido pom per Lye Mawr \e iE hao teivoy a. J Page rs —— ii Ting Senin date? [Segptufre vn aa of __|| Time aeceo date & & data which 15 Aeguenced olaba willie Spot Srna pouiod tke ola OZ : . This data & U a Ce rad pet | Time Leven ol ta | a meadumonts uhh fork place Ab AN Rontant Cnbaval y time . | Dt He 9 i Aga dat So eso — enampe —edate tempera tate chang. hourly é in INK = UD rate dacly SSO pile malty hema a, | Se he. ae f Wisteie twcke ANN: 2 Undtabte adironk se Gene nck A — tng fy padi wilted ot «ach tema plaligs adoont Ea — fle Use ormelle lass goo ha Lan ntl a bed lon trav pees Layer p buku “eke bammpo® Sons Exuiren unt 2 str wre Miele join 2 Sub YIU agen pponr? —_ Q 9 qoed aptims Ak Senatded. bey coun a > wh ahi y Kuali wt ANd bey overt | Pobicy 9 Ot bp Ste &% sul ee gent u . felines lei ing F classmate 1 EB | OPicinneed = 8 hae fre mayfmun. Wiatg ital Aidtauee . Riahgtnal vanes apprised by okedd Ln Ls parallel bo Mapes. 7 4 _yectors ~ thier at fhe data pom — fH tant tothe le whig or the nos. Shot vectay uw | ‘ ee sete aad veeoe, pe oN pti Mappare 1 || &m Som ve wae Merge Lay a0 ou [i anc | OO Tt sum yp Su Tt LY Areas Symi 2) ten Lineae Svm Warton sym fit [s wasd for Lintarly scpora ele ad P thot Mtouy e « dafoct tou ke wlnwitied Maid 2 Shas r hune Hom tats dala jr : a Ly ae Non Aenea Sum - tS veel for pons Le atpraba data utich ecomnot be uate A Biraghe Line . Y es Ae Roaslion \p called non fthea SVM a 1 ye hitafe

You might also like