0% found this document useful (0 votes)
189 views32 pagesML Unit-1
The document provides an overview of machine learning, including its types such as supervised, unsupervised, and reinforcement learning, along with their respective algorithms and applications. It discusses the challenges faced in machine learning and highlights the significance of deep learning as a subset of machine learning, emphasizing its capabilities in handling complex patterns and large datasets. Additionally, the document outlines the advantages and disadvantages of artificial intelligence and machine learning in various fields like finance, healthcare, and engineering.
Uploaded by
nagasuri lavanyaCopyright
© © 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
0% found this document useful (0 votes)
189 views32 pagesML Unit-1
The document provides an overview of machine learning, including its types such as supervised, unsupervised, and reinforcement learning, along with their respective algorithms and applications. It discusses the challenges faced in machine learning and highlights the significance of deep learning as a subset of machine learning, emphasizing its capabilities in handling complex patterns and large datasets. Additionally, the document outlines the advantages and disadvantages of artificial intelligence and machine learning in various fields like finance, healthcare, and engineering.
Uploaded by
nagasuri lavanyaCopyright
© © 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
/ 32
Machine Leayninq -
| Uoit-1
hhoduction: Artificial! Shteltiqence, Machine learning, Deep
| learning / Types of Machine leaxning systems, Main challenges
| of Machine eon
Statistical Keorning: Thbroduction, Supervised 4 unsupervised
| learning , Fini 4 test 1088 tradeoffs, in statistical tearing,
| Estimadi yisk Sotistics, Sampling distribution 4 an
fae empirical isk minimization -
Supewisedl Learting: Basie method : Distance bared methedt,
| Neorat neighbours, Decision treet Aleive Bayes
| Lineay models: Linear regreation logistic regredion, generalized
Vinear modelt, Support vector mochinel -
Binwy clayidication. Multiclay /chuctec! outputs , Mls Rocking.
Unit-3
Ensemble learning + Rendom tovets: Dnbroduction, votin.
clomiter, Bogging & ‘porting, Random Foret, Busting, stacktng :
Support Vector Machine: Linear SVM claification , Non-linear
Sym clomification, SYM regression, None Bayer clouifiers
unit-& ‘ .
Unsupavised Learning Technique: clustering, k-means, limit y
Kerneang, Wing clustering for tnoge Seqinentation, sing
clustering tov preprocettig » using clusteting for Sern}- supervised
learning, DRSCAN Gatunian minturel-
Dirmensionalily Reduction: The curse q dimensionality, Vain
Approocher fox Dimensionality Reduction, Pen, wing, Scikit~
learn, Rardamized pen, Kernel pcr
Unik 5 ri "
at Netusorks 4 Deep Learning: Sntreduetion ty Artrticial
Neurol pletvoorks with Kers, implementin MiPs vorth kes ,
Srotalling “Tensorftou2, Loading & preproeey ing data with
Greting: ae
ONT
=| Explain the ollosing AP sees '
¥ Pybficial Thtelegence tap
aDeg Leasning bese | ack
3 Machine Leasning (ML) oe
T yebes 4p the Avifical Aleligee tohich is ‘it
Prpted by Tobn Macarthy in 195%.
HL is sub divided into different types-.
Rule Bored as ae
a Contert based AI .
“8 'Ged “BP 2x : .
A AL is sub divided ito “subs branches, Tike. deep
leaning ‘Machine teanting’ dnd; neural, networks: .
OR AL requires a special harduoare : ond’ solwoare.; for,
| toring and haining 'ML allgarthms sa
@ AS supports the Tollowing programing, larga like
|" python, R- programing Tava FetOT 1
fp AI depends on the Following 3 eolre
| wean
W Reasonin'
| Gi self Cov
4 Al. progratrimes mainly gathers data, ereoding rulel, evecurti
the a and thee 48 axe known at algorithms: These
algorithm, axe executed step by step 10 achieve the
| speciic task :
ct Advantages! once
‘
oy Good ‘at dealing’ renter jos score prin)
“ha Me yeduces* time: >: _
ca ‘Ot completes humap- np, douke-
Disodvantogas ee
# Expensive ae 3
+ Limited qualified workers 0 bt “Al sili! Mt hock 4 pbility
* Requirer more deep Keoxning techniques
Appear:
& Medical Field
H Science & ‘econly
¥ finance
* Ashobogy
# Engineering ete.
Sports, entdloinmen
+ News ete:
‘| Deep Leorring’ oe ee
A Deep learning is a brorch of machine leeyning baed of
Arlificial Neural” nfetuoorkt *
Deep ein is capable eg complex pallens 4
yelationships “within dato : Funetionalitiey
.| Rothe neutral nfws ove inped y “the. struetores of urn
“| brgin's § biological newwons) |
* Deep leanning algun exe axe os to learn fom large
arnounts eae, iz . .
-
r6 WPiilagerl
a ooh
“t ft etlayer/MARey toe
Tally” connected init layers oe
deep, les learn ing. alge algo vithms can oatomatically learn ivnprove
Ext nad the: need oof manual tecture engineering
om dato. "ws
$ deep leavring hay ackievee! more -sucten jn vorious fields
Tike imoge recegnisétitn, /nafural’ Langtiage: prep Proceuing, speech
yecog Mitinny reeomendation sytem, A\ cars ete:
$hee one 3 type neural netwoor:
1 CAIN's = Convelutional. neural networks»
a RNAS ~ Recurrent Neural Networks: . ~ 3 DBN's- Deep belief Networlu: : Iya} .
lt To ‘oi deep: neural nls, we, we oO Specialized leo
high contention —LhL Siig int ou
Ht Deep heorning can be used For supervised, anguperised feeming
4, Reinforcement Learning. 1 suid
Katong of Neural RABE i
ae the inputs. ove given, tothe input layer which is the
Fisk layer of ANAL .
* The inpudt are Poxed on to the hidden layers which tt
Storts from the Second layer.
# Each neuron is given a volue known a4, weights.
$ The ‘hidden toyer gek the infomation from the neurons in
ahe previous layer 4 computer the weighted total'z “then :
tronsferg it to the newont’ in the nett layer , \
F There voeights axe adjusted curing he chaining proces, fo
improve the perfomonte of. the models
# Finally the efp is pramt at the ofp layer cohich is “he lost
layer of ANNE
Applications = : a
Fhinonciol sewite
3 Medical Fields [health cone Services ~
x Froud detection. ; "OU
Fe agonal
* penicechy eX? sii
|g Zeustomer’ Service!
% Law Enforcement
@ |(Mochine Leasing: |
' pawntaqa* | i |
Fheeurate resulls can be obtained by deep leering ‘lgpitns
| DH leorns From huge amounk of data:
Ditodvontoge!* /
evey complex in structure |
Hfective (expensive) /
se dime taking proce tor learnin
$ vend difficult to Find the /-frouble cheat the err neuen
FANCY Gee Linke are reguired ett: : 3 Vachine Learnings wl 1 essa
(* Machine leaning & a braneki gf. PrUleiah Istelhigene
ie ML develops clit by leoaring the hidden’ pattemy:
ie dota (Stienee) sels used to make preclietiong oN Si Y |
Type dota woitheut tising pregame Joh each fark |
%.ML 18 ted in sary diFfevent Application, Like ina 4
Speech * recognition, noduial Language proceiting, Fraud det tly
Gutomemow vehicle, clones, robots rede 5 irae
flowy” ML’ aloottfin coorks:- aN :
* Meweard Pawes Sh this step the ML; algailbm take -
1» | MpUt data, produce, an olelpul: predictions owe’ made
"| depending or -the mactels 9. Ve of wien |
& Lose Fuungtion’ The: loss: Function «is :aléo knows’ ox enor'or |
Gost function: Tis used to’ ealuate the accurauy of he
Precictiont made by the tnodel: “The fuxietion’ com pares ¢ |
Predicted’ cotpul with “the actual’ outpot: This difference |
| & known ay emer or bus the goal - he ‘model isto |
| yecluce the eros 10 increase Be perlormnorice’ af the mecel
[2 Model eplindzalion pagys His on ite’ proto: jig
intemal parameters gf ty imel to miniinize the error 01 “low
[Function “This. is knouon’ 04 optimisation process: we: Ute an
optimization al me like gpodient descent The algorithm
repeats athe proten of adjusting internal parameters to reduce
Abe evror and Ao intreate “the Accmvocy« , ‘
ML tie Cyaes ,
|b Study he problems
1) & Data ‘collection
3. pala pr cation ‘
yy Model Selection i
5. Model kuilding & haining
§ Vodel Evaluation
+ Model Tuning 0"
Luge Deployment’ ‘ ve t if
a Monitoring & Maintainancess tokt pesueir antes P| Typo of MI »
Atha ae 3 Aypet gy ML
br Supewised ML
4 Unsupervised, ML
3-Reinforced Mp
Supewised Mis 7
A Supervised Learning. is a ML In echich the alga
11S tained. on lated fated se ih the
ED Learns to Map input Jeatuves to the targa cufputt based
00 — labeled ‘hahing dala” OT Caran 3d
#O Supervised ledvning , the algorithm iS provided! with inpute
4 Concspending output labeli- And it’ leams 46 make prediction
OM new unséen data :
& There Gre & typer .of Supervised Leavting
b Regression + TES Ly 4 ‘Superviced * Ledaning where ‘the
algorithm learns 40 predict values basedl on inpubt: The -otrtput toi
Reqyeuion algorithm hookt like Yet/noy ‘eft hue/Falte ete:
4 DiHeent type’ df Yeqrewion “algosthns axe’ Linear regresion
| polyromiat algorikm) dedision dree regression ridge regreation,
Pardo Fovet regression; Suppot - vector regpession ete:
Application: Banking sector, 1 |
a alock exd ge ‘phic heuig pricel, Petroleum pricet ett.
a clan ication’ 9k isa ype. 4: Supewised Learning cohere
th algorithm learns! 40 as a agg calegory. bed on
inputs: The outpat of clanifcatinn algarifbm; ove diserele
Yohier: ctawsfication “algoiifyms > be binary oF rnulticlass the
ditfaat claniticotion” sits ore’ logistic re pestion, nalve-boved,
} decision heer, SUM, ke necned neighkouy algoribm( kan)
unsupewited ML + sh Takacs
# vowpenised Leaning. Iso dupe, af Mb where she algiibn
Hearne fo recognize the patterns. ip data, without bing
explicitly srorned. ving Labeled data '
# the qal ugupited heawing is fo digeover the under
Wun (Raden edetuve* ot data.” ys A thee axe Mainly a typel J unsupervised’ leavning
Dclusteing - chasteving algorithms Quoups similar cata’ pring
| dogelhey bated on their dhavactastiel * i
bthe gal is 10 Uertify groups 0» clustert of ta point
|dhot axe similar to each other j
¥ Each cluster is digtinet to other clutter: clustering algoritn,
ove Kemeant, hiearchical eluterial 4 DBSCON: ;
W Dimensionabty te duction:- Dimervionally reduction algorithms
Ne Re /
Yeduee the input variabler in this data et il presen
original infomation os powible: ai
$s ie welul for reducing she commpleuily of datoxel
(3k allo maker eatin to vicualize & obalyze |
ir some Popularity Dimensionality reduction ogoithns ave(pep)
® R la mai Analyt, t-sKl€, autoencedlers ete:
| Kanrorcemer °
+ Reintorcenent “Teosting tiga ype of ML tohere an ae
leoang to. interact woth an envionment by pettorming « j
e receiving revoards ard penalities bared on i actions.
% The goal of reintcerment heosring 1s to hear a policy which
| maps from du 4o actions evi
| i ere ave monly a Aypa 4 “Reinforcement hearing
I \ Model bored Reinforcement Leaxning cardyetvah
| | a-Medel free Persforcement Learning’ wuse4 suck
|T Nadel ad Renfeament Laut’ Yo his mel the
agent Jeoa a/ roodel f envivonrnent. with Similar stata, +
rewards associated with state ‘action Pair (é: iy as aig
[ie satity the requirements like, clearing the jelly will get
- | Yeward) a
atthe agent user the model to masimize the expected reioarl
% Come populor enodel bored Reinforcement leaming algorithms
axe value-iteroction alguribm, policy - iteration algoitbmn :
[8 Model Tyee Reinforcement heorning- Dn: This: model sthe
| agent leouns o, model directly fiom’ the experience: Algitbme aye ted
Abe hidden pattemt + relationships
Suriers, PUBG Temple
(subi
ent interacts with th
% The ag
poli
Pts environment) :
& Some lox model Free Rei
axe @- learning. alqorthn, sa
yeinforcement heeaning f alg ovithn
beued on the vewards it
mn) + ”
2 envienment 4 updates its
receWver.( changing the Suiony
inforcement Leami alocnithmy
RSA algnilhm, and’ deep
Differentiate. bl machine Learning 4 deep leaming -
Machine leasing _ f
Ho leon
in she datoyets:
Thee algorithys can’ utork on
gmoller amounts af dada sel.
gy There algorithms’ ave better
Foy how Labelle! tork:
gy Thae algorithru fake len”
Yime to train the model. 7
Flos complex A OL Ay
inkapet the vault:
3 Thae “algorithms ton woik on
tpu yequirer lew computing
er
$1 model is created by wu
Yelevont feature! which axe!
roonually ehocted trom the,
image raidprock”
Deep \earnin
F9E igen artificial Yeosrning)
neural {ANAL A yehitecture fo
learn ~the ; :
FTE vequiver Loge volume
doteselt compared’ to. ML
A thee algorithm: axe better
for comple fosty like i
priceiting, natural langoage
procening ‘eles ovr
Lthae” algorithms faker more
die to tidin the mode!
F Move complex A it works like
\.{ black bot intepretotion:
e yequlh 1s nob Caay.
OF require high perfomance
computer” with. po
LEA rnedel te, created by a
yelevont feotuver which axe
latorrotically extiacted trom
athe image: “Ot ig an end to
* | ead procen- \
\ List, out Voriou
Supevised leaming
+ Thee algorithms aye -hained
by uiing “Labelled date
¥ Supenise d: learnh _ modell
take direct Feedback to check
iE ik is conect ofp or not
SSupewised Learnt rmoctell
Predict the olp-
#9 slerted leasning “ilp
dota it provided to the model
along with the olp |
rs is, wed for, prediction
¥ VM is divided into
» \Regretion aly
Lclauiticetion
FAN) layer ove known
$b viet offline analisi 4
doto
tr ienage reshma
devoice or Spece vrefoginition
nsupense) Leming
X Thee algonitins, ane toing|
using uolakellel data.
¥ Unsupervised heaming Model
doen take any Feedback -
Mt finols the hidden |
Herne in dota:
4 Do sthis only ilp, ddto.
18, prided 40 the model
4D ig ued tov Analyis
gk is divided. ints a typa
clustering
q. Awociation
zAN cloud are unknown
ee
Appheations of ML
PV wa realtime Analysis
4 dato
> 3. Tratfic diction’,
us Prod qeroromendation
1g, Stk diving ¢axt
ge e-mail sparn 4 maloore Filtering .:
vitval pextonal “auittante < like Alera)Siti efe «
g-Stock’ market -hading:
g-Aulomatic harguage transition
Gage Reps
# nage Retagnitton 1s the one of -the most Common
Applications ot ML
& % is used +0 identity the objeds pesont, Placa, digital
Sa inmga ete: 2 popula» Application of Simage Recognition 4tace detection iss
“qutornated Syiend 0gqing Suggettion Carys), Jacebook. .provicler
[4here appl ation. dov 4.agging Friend ar
gumbockiag 0 phone eorth image Recognition *
“Woice_ov Speech Recognition 2 :
3 coon provides an option of "Gearch by voice”,
\a G9 fe qusistant , sini, cortanay lena ave HINGE speech
|yecegn’ Hon techrehoy Yor yoriouy application *
|% speech yecognition i$ a proce Lhat converting. voice into
Hext. 4 it is alto known 08 speech 40 text * }
Frattc predition:
“Ab is wed, in Gocgh raps aot
“Ris shows the cont ard shalt poth with the traffic
(conditions - Blue Coled poth leu Aralfic A ved colored posh
pdicter lente. 01, Nigh: -bedfie. poth + it allo predict the
dime fron Source to destinations t
“Product Retanmendations oom 4
AML is widely wed by, e-commerce ofphectiont + eplertainment
ry like Thipkasts, Amazon Netflix eles ‘
& Product reorrenendection ML algoritters depenolt upon the yarn
previoud” browsing history of upy, diffoat codegorics (items)
‘histoy of 'semched : by the wey, Annual espera 4 the
ee
athe amount. of the time: sperd by, the Laer. on the app -
yp Netti: Bho, Prime Hotttor sugges -the ryecommendlotion Fos
entertainment erie Movies seb oe ,
Sa diving Care, PPP bar odaenssoe i ,
3 Most of the exdting application aye elt. driving Cavs:
3 Tale cor is the west popu con with ouopilot
ee wer Uriupervided leaxning 40, drain the Car sto detect, the
|persons, objects: ‘ \
“Bra spam ard) Mahoave Filtering :
limes Spam Filers ore wed by Ana
applica tio X When we receive a mail. itis aectomaticaly detected a,
indportant oy) novinal’ or span | balow
SPAM malt are rosted to
