You are on page 1of 10

BasicsofMachineLearning

Seshan Srirangarajan
Dept.ofElectricalEngineering
IITDelhi

Outline

Machinelearningandthelearningproblem
Prerequisitesforlearning
Everydayexamplesoflearning
Componentsofmachinelearning
Typesoflearning
Somemoreexamples

MachineLearning
Systematicstudyofalgorithmsandsystems
thatimprovetheirknowledgeorperformance
withexperience
Verysimilartohumanlearning

Prerequisitesforlearning
Patternmustexist
Patterncannotbedescribedmathematically
Availabilityofdata

ExamplesofLearning
Everydayexamples
Spamfilterinemailapplications
Recommendationsorsuggestionsofferedon
onlineshoppingwebsites
Loanapprovalbybanks
Insuranceriskevaluation

ComponentsofLearning

Inputdata(x)
Outputortarget(y)
Targetfunctionormapping:{f:X>Y}
Data:(xi,yi)fori =1toN
Model:g:X>Y(gtriestoapproximatef)
Inlearningproblems,targetfunctionfis
alwaysunknown

Unknowntargetfunction
f:X>Y

Trainingdata
(xi,yi),i =1toN

Learning
Algorithm
SetofModels
Learningmodel

FinalModel
gf

Learningtoaddressatask
Task

Domain
objects

Features

Data

Trainingdata

Learningproblem

Model(g)

Learning
Algorithm

Output

TypesofLearning
Supervisedlearning
Inputdataislabelled:(Input,Targetoutput)

Unsupervisedlearning
Inputdataisunlabelled:(Input,??)

Semisupervisedlearning
Reinforcementlearning
(Input,someoutput,rewardforthisoutput)

Examples
Whattypeoflearningisbestsuitedforthe
followingscenarios
Coinclassificationsystemwithcoinspecification
dataobtainedfromtheMint
Coinclassificationsystemwithalargesetof
labelled coins
Acomputerdevelopingstrategytoplaychessby
playingrepeatedlyandadjustingitsstrategyby
penalising movesthatleadtolosing

You might also like