機器學(xué)習(xí)分析報告

ML-MACHINE

LEARNINGDL-DEEP

LEARNINGAI

–

ARTIFICIAL

INTELLIGENCEDS

–

DATA

SCIENCEVENKATA

REDDY

KONASANIPART-1What

is

MachineLearning?2ActivityC

l

o

s

e

y

o

u

r

e

y

e

s

a

n

d

th

i

n

k

a

b

o

u

t

tw

o

te

r

m

s

–M

a

c

h

i

n

e

L

e

a

r

n

i

n

g

a

n

d

A

r

t

i

f

i

c

i

a

l

I

n

te

l

l

i

g

e

n

c

e

.What

is

the

first

thing

that

comes

toyour

mind

when

you

hear

thesetermsMachine

LearningArtificial

Intelligence3WHAT

ARE

YOU

THINKINGABOUT

–

ROBOTS?4OR

ANY

AI

BASED

MOVIE?5HAVE

YOU

THOUGHTABOUT6MathematicsStatisticsDatasetsData

AnalysisOptimizationAlgorithmsData

MiningMachine

learning

(ML)

is

the

scientificstudy

ofalgorithms

and

statistical

models

that

computersystems

use

to

effectively

perform

a

specific

taskwithout

using

explicit

instructions,

relyingon

modelsand

inferenceinstead.OK

GOOGLE

…WHAT

ISMACHINE

LEARNING?7WHATREALLY

IS

MACHINE

LEARNING

?-

WIKIPEDIAUsing

historical

data

to

make

futurepredictionsBuilding

models

on

historical

data

topredictionsTaking

training

data,

building

models

onthe

training

datausing

themodels

tomake

the

future

predictionsMaking

themachine

learn

the

patternsin

the

data9IN

SIMPLE

TERMS

..DATA

IS

IN

DIFFERENT

FORMSNumerical

dataImage

data

(pixel

intensities)Video

data(frames

per

second)Sounddata

(waves)Text

data

(tweets,

comments,

feedback)10APPLICATIONS

OF

NUMERICAL

DATACREDIT

RISK

MODELSIdentifying

risky

customersbefore

offering

a

loan11MARKETING

ANALYTICSDo

you

receive

any

marketingcalls?

Have

you

ever

receivedany

marketing

call

for

Audi

car?RETAIL

SALES

ANALYTICSHave

you

ever

wondered,

whyonly

you

are

gettingpromotional

offers

on

clothsand

accessories

where

as

I

amgetting

offers

on

apartments?FRAUD

ANALYTICSHow

does

a

bankdecide

thepotential

fraud

transactionsfrom

millions

of

credit

cardswipes?Face

recognition

–

Using

image

as

input

dataObject

recognition

–

Pixels

is

the

input

dataDigit

recognition

–

Using

text

as

imageSelf

Driving

Cars

–

Using

video

data

as

inputAPPLICATIONS

OFMACHINELEARNING

–IMAGES

AND

VIDEO

DATA12IMAGE

DATA

IS

ALSO

NUMERICAL

DATAS13Image

dataHuman

VisionComputer

Vision-1-1-1-1-1-1-1-10.9-0-1-1-1-1-1-1-1-1-1-1-1-1-10.310.3-1-1-1-1-1-1-1-1-1-1-1-1-011-1-1-1-1-1-1-1-1-1-1-1-1-10.810.6-1-1-1-1-1-1-1-1-1-1-1-10.510.8-1-1-1-1-1-1-1-1-1-1-1-10.110.9-0-1-1-1-1-1-1-1-1-1-1-1-011-0-1-1-1-1-1-1-1-1-1-1-1-10.910.3-1-1-1-10.510.90.1-1-1-1-10.310.9-1-1-10.111111-1-1-1-10.810.3-1-10.410.7-0-011-1-1-1-1110.1-10.110.3-1-1-010.6-1-1-1-1110.80.310.7-1-1-10.510-1-1-1-10.811110.50.20.80.810.9-1-1-1-1-1-00.8111111110.1-1-1-1-1-1-1-00.81111110.2-1-1-1-1-1-1-1-1-1-00.30.810.5-0-1-1-1-1-1Sentiment

AnalysisExtraction

of

key

topics

in

the

dataDocument

ClassificationAPPLICATIONS

ON

TEXTDATA14PART-2What

is

Deep

Learning?15ANNANN-

Artificial

Neural

NetworkANN

is

oneof

the

technique

in

Machine

LearningANN

has

input

layer

,

hidden

layerand

output

layerFor

a

really

complex

and

nonliner

datasets

we

need

several

hiddenlayersANN

with

multiple

hidden

layers

is

known

as

deep

neural

network16ANN

with

a

single

layer

isknown

asshallow

networkANN

with

multiple

hidden

layers

is

known

as

deep

neural

networkNot

just

multiple

hiddenlayerssometimes

the

type

of

hiddenlayer

isalso

different.This

concept

of

solving

problems

with

multiplehidden

layers

is

knownas

deep

learning17DEEP

LEARNINGDEEP

VS

SHALLOW

NETWORKSA

neural

network

with

single

hidden

layeris

called

a

shallow

networkA

neural

network

with

more

than

onehidden

layer

is

called

deep

neuralnetworkshallow

networkDeep

network18DEEP

VS

SHALLOW

NETWORKSA

singlelayer

might

not

have

theflexibility

to

capture

all

the

non

linearpatterns

in

the

datashallow

networkDeep

network19A

deep

network

first

learnsthe

primitive

features

followed

by

high

level

features.

This

helps

in

building

efficient

modelsLot

of

experiments

have

shown

that

a

deep

network

with

lessparametersperforms

better

than

ashallow

networkFor

example

deep

network

with

hidden

nodes

[10,10,10,10]

mightperform

better

than

shallownetwork

with

[80]hidden

nodesDeep

neural

networks

are

amazingly

powerful.With

sufficient

number

of

hidden

layers

and

nodes,

we

can

fit

a

modelto

any

type

of

dataThey

have

the

power

to

capture

any

amount

of

non

linearity20DEEP

NEURAL

NETWORKSDEEP

LEARNINGIS

A

SUBSET

OF

MACHINE

LEARNINGMachine

LearningDeep

Learning21PART-3What

is

Artificial

Intelligence?22MACHINE

LEARNING

MODELSTraining

dataBuild

Model23MACHINE

LEARNING

MODELSNew

dataApply

ModelClass1GetPredictionThis

prediction

can

beright

or

wrong24MACHINE

LEARNING

MODELSNew

dataApply

ModelClass2GetPredictionOne

way

modelsThis

prediction

can

beright

or

wrong25AI

=MACHINE

LEARNING

MODELS

+FEEDBACK

LOOPTraining

dataModel26AI

=MACHINE

LEARNING

MODELS

+FEEDBACK

LOOPNew

dataApply

ModelClass2GetPredictionFeedback

Loop27AI

=MACHINE

LEARNING

MODELS

+FEEDBACK

LOOPUpdate

Trainingdata

based

onfeedbackUpdate

the

Modelbased

ondataPredictionFeedback

LoopClass228Manual

entry

after

goingthrough

testcases

–

Google

mapsIndirect

feedback

collection

based

onuser

actions

for

-

User

click

vs

notclick

on

your

YouTube

adIndirect

feedback

collection

based

onactions

–

In

case

of

self

driving

car,hitting

a

wall

is

an

action.HOW

IS

FEEDBACKCOLLECTED29Self

driving

carsSIRI

/

Ok-googleAlexa

/Google

homeRecommendation

systemsImage

recognitionSpeech

recognitionSpam

filteringAPPLICATIONS

OF

AI30MACHINE

LEARNING

IS

A

SUBSET

OF

ARTIFICIALINTELLIGENCEArtificial

IntelligenceMachine

LearningDeep

Learning31PART-4What

is

Data

Science?32Data

Driven

Decision

makingMaking

sense

out

of

dataFinding

hidden

patterns

in

the

dataAnalysis

using

not

just

machinelearning

models

but

also

using

datavisualizations,

intelligent

reportsMost

of

the

techniques

and

toolsseen

indata

analysis

in

early

days

arenow

falling

under

data

scienceWHAT

IS

DATASCIENCE?33MathematicsStatisticsCodingDatabase

managementData

AnalyticsPredictive

modellingMachine

LearningDeep

LearningDATA

SCIENCE

IS

AFUSION

OF

MANYFIELDS34DATA

SCIENCE–

FOUR

MAJOR

TYPE

OF

SKILLSDatabaseAnalytics

&MLBigdataPresentation35THE

TECHNIQUES

YOU

NEED

TOKNOWDatabaseKnowledgeData

base

ManagementData

blendingQueryingData

manipulationsETLPredictive

Analytics&

MLBasic

descriptivestatisticsAdvanced

analyticsPredictive

modelingMachineLearningBig

Data

knowledgeDistributedComputingBig

Data

analyticsUnstructured

dataanalysisPresentation

SkillData

visualizationsReportdesignInsights

presentation36MACHINE

LEARNING

TOOLS

AND

SOFTWARE'SDatabase

toolsSQL/MySqlOLAP

cubesTeradataDB2/Sql

Server/

Oracle/Informix/ExadataAnalytical

toolsSAS/R/SPSS/PythonWeka/MATLAB/TensorFlow/OCRBig

Data

ToolsHadoop,

Hive,

Pig,Mahout,

Spark,

JavaPresentation

ToolsExcelTableau,

Qlikview37DATA

SCIENCE

-DESIGNATIONSDatabase

DeveloperETL

DeveloperMIS

&

DBDeveloperData

ArchitectData

EngineerData

AnalystStatisticiansBusiness

AnalystData

ScientistBigdata

DeveloperHadoop

DeveloperSoftwareEngineerMIS

AnalystReporting

AnalystBusiness

Analyst38Data

ScienceMACHINE

LEARNING

IS

A

PART

OF

DATA

SCIENCE39SArtificial

IntelligenceMachine

Learning*

These

are

individual

interpretationsDeep

LearningPART-5The

Learning

Path40FAQ

BY

DATA

SCIENCEASPIRANTSI

want

to

be

data

scientist

whattraining

should

I

take?I

already

have

knowledge

on

fewtools,

what

are

my

next

steps?What

skill

should

I

add

to

my

profileto

make

it

to

next

level?I

am

new

to

data

science,

where

can

Istart

?41You

need

training

basedon

your

skill

level.Based

on

skill

set

we

can

divide

the

whole

datascience

aspirants

into

four

categoriesBeginner

-

Completely

new

to

Data

Scienceand

MLIntermediate

-

MIS

and

Reporting

AnalystAdvanced

–

Data

Analystand

PredictiveModelerComplete

Data

Scientist

–

ML,

Hadoop,

R,Python,DL,

AICATEGORIES

OF

PROFILES42THE

LEARNING

PATHS

43Tools

&

CodingR/SAS/Python/Hadoop/WekaBasic

Statistics

andMathematicsBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessTHE

LEARNING

PATH

–

OUR

SUGGESTIONSS

44Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessDo

not

try

to

learn

all

the

steps

in

one

sitting.You

need

to

learn,

absorb

and

then

practisebefore

youreach

the

next

stepMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1THE

LEARNING

PATH

–

OUR

SUGGESTIONSS

45Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessR

or

Python.

Bothare

really

good.

Pick

any

one

of

themIt

also

depends

on

your

business

problemIf

youareplanning

to

learn

deep

learningthen

go

for

pythonMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1THE

LEARNING

PATH

–

OUR

SUGGESTIONSS

46Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessDo

not

start

with

stage-2

or

stage-3

directly.Strong

fundamentals

will

make

thelearningeasy

in

later

stages.MathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1THE

LEARNING

PATH

–

OUR

SUGGESTIONSS

47Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessWhile

learningtheseconcepts,

try

toavoid

academic

style

courses.Look

for

the

courses

with

lot

of

hands-on

exercises

and

case

studiesMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1THE

LEARNING

PATH

–

OUR

SUGGESTIONSS

48Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessDo

not

focus

on

the

tool,

focus

on

the

technique

and

algorithmLearning

python

or

R

tool,

will

not

makeyou

a

datascientistMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1PART-6Course

Curriculum49FOCUS

IS

ON

FIRST

TWO

STAGESS

50Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1DURATION

–

10

DAYSS

51Basic

StatisticsandBasic

Algorithms

-Regression,Classification

andSegmentationAdvanced

MLAlgorithms

-NeuralNetworks,

SVMs,Random

Forest

andBoostingDeep

LearningModelsCNN,

RNN

and

LSTMAIModelsDeep

Q

LearningReinforced

LearningMarkovDecisionprocessMathematicsTools

&

CodingR/SAS/Python/Hadoop/WekaStage-2Stage-3Stage-1TWO

PHASES52PHASE-1

(5DAYS)Python

for

data

scienceData

manipulationsin

pythonBasic

StatisticsData

validation

and

CleaningRegressionLogistic

RegressionDecision

TreesCluster

AnalysisModel

Selection

and

CrossvalidationANN

–

Artificial

Neural

networksSVM

–Support

Vector

MachinesRandom

ForestBoostingNLP

&

Text

miningTensorFlow

&

kerasDeepLearning

ModelsConvolution

Neural

NetworkRecurrent

Neural

NetworksPHASE-2

(5DAYS)100%

Hands-on

Training30

case

studies

laced

in

the

courseCreated

for

Non-

StatisticiansDatasets

from

multiple

domains,codes

files

and

in

class

exercisesTeam

assignments

and

mentoringFinal

AssessmentE-learning

material

supportCOURSE

FEATURES53PART-7Data

Science

and

Machine

Learning

MythsSMYTH-1

:

MATHEMATICSMyth-1

:

To

be

a

gooddata

scientist,you

need

to

beexceptional

atstatistics,mathematics,calculous,

algorithms

etc.,Not

necessarily.55SMYTH-2

:

PROGRAMMINGMyth-2

:

To

be

a

gooddata

scientist,you

need

to

have

exceptional

codingskillslikePython,

Java,

C++

etc.,Not

necessarily.56SMYTH-3

:

COMPLICATED

MODELSMyth-3

:Data

science

is

all

about

building

complex

predictiveandmachine

learning

models

to

solving

business

problemsNot

necessarily.57SMYTH-4

: