Machine Learning Ch 1.ppt

DBATU

1. INTRODUCTION
TO
MACHINE LEARNING
CHAPTER 1

2. Topics Covered
1.1 Introduction to Machine Learning
Artificial Intelligence
Machine Learning
Application of Machine Learning
1.2 Types of Machine Learning
1.3 Supervised Machine Learning
1.3.1 Classification
1.4 Unsupervised Machine Learning and its Application
1.4.1 Difference between Supervised and Unsupervised Machine
Learning
1.5 Semi-Supervised Machine Learning
1.6 Reinforcement Machine Learning and its Application
1.7 Hypothesis Space and Inductive Bias
1.8 Underfittingand Overfitting
1.9 Evaluation and Sampling Methods
1.9.1 Regression Metrics
1.9.2 Classification Metrics
1.10 Training and Test Dataset and Need of
Cross Validation
1.11 Linear Regression
1.111 Linear Models
1.12 Decision Trees
1.12.1 The Decision Tree Learning Algorithm
1.12.2 Entropy
1.12.3 Information Gain
1.124 Impurity Measures
Exercise

3. Introduction to Machine Learning
Machinelearningisabranchofartificialintelligence(AI)andcomputersciencewhichfocusesontheuseof
dataandalgorithmstoimitatethewaythathumanslearn,graduallyimprovingitsaccuracy.
MachineLearningisanumbrellatermusedtodescribeavarietyofdifferenttoolsandtechniqueswhich
allowamachineoracomputerprogramtolearnandimproveovertime.
MLtoolsandtechniquesincludebutarenotlimitedtoStatisticalReasoning,DataMining,Mathematicsand
Programming.
Tolearn,amachineneedsData,ProcessingPower/PerformanceandTime.Itcouldbesaidthatifamachine
getsbetteratsomethingovertimeandimprovesitsperformanceasmoredataisacquired,thenthis
machineissaidtobelearningandwecouldcallthisprocessMachineLearning.

4. Introduction to Machine Learning
Machines/computersanabilitytolearnthewayhumansdo,i.e.withoutexplicitlytellingthemwhattodo.
Machinelearninggivescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed.
ArthurSamuel
Machinelearningreferstoteachingdevicestolearninformationgiventoadatasetwithoutmanualhuman
interference.

5. Well Posed Learning Problem
Awell-posedlearningproblemisataskinwhichtheInput,Output,andLearningobjectiveareclearlydefined,andthereexistsa
uniquesolutiontotheproblem.
Awell-posedlearningproblemhasthreeproperties:
1.Existence:Theproblemmusthaveatleastonesolution.Theremustbeapossiblerelationshipbetweentheinputandoutputdata.
2.Uniqueness:Theproblemmusthaveauniquesolution.Theremustbeonlyonecorrectrelationshipbetweentheinputandoutput
data.
3.Stability:Thesolutiontotheproblemmustbestablewithrespecttosmallchangesintheinputdata.Theoutputproducedbythe
machinelearningalgorithmshouldnotchangesignificantlywhentheinputdataisslightlymodified.
4.Awell-posedlearningproblemisessentialforthedevelopmentofeffectiveandreliablemachinelearningalgorithms.Withouta
well-posedproblem,thealgorithmmayproduceincorrectorunstableresults,makingitdifficulttouseinpracticalapplications.
Soitisimportanttocarefullydefinetheinput,output,andlearningobjectivewhenformulatingamachinelearning
problem.

6. Well Posed Learning Problem
Alearningproblemcanbedefinedasataskinwhichanagent(suchasAMachineLearning
AlgorithmoraHuman)mustlearntoperformaspecifictaskormakepredictionsbasedonasetof
inputsordata.
Threefeaturesthatcanbeidentifiedinalearningproblemare:
Inputdata:Thisreferstothesetofdataorinformationthattheagentusestolearnandmake
predictions.Theinputdatacanbestructuredorunstructured,andmaycomefromavarietyofsources
suchastext,images,audio,orsensordata.
Outputorprediction:Thisreferstothetaskthattheagentistryingtolearnorthepredictionthatitis
tryingtomakebasedontheinputdata.Theoutputcanbeasinglevalue,asetofvalues,ora
probabilitydistributionoverpossibleoutcomes.

7. Well Posed Learning Problem
Evaluationmetric/Performancemeasure:Thisreferstothemeasureormetricthatisusedtoevaluate
theperformanceoftheagentonthelearningtask.
Theevaluationmetricmayvarydependingonthespecificlearningproblemandmayincludemetricssuchas
Accuracy,Precision,Recall,F1Score,orMeanSquaredError.
Definition:-
AcomputerprogramissaidtolearnfromexperienceEwithrespecttosomeclassoftasksTand
performancemeasureP,ifitsperformanceattasksinT,asmeasuredbyP,improveswithexperienceE.
TomMitchell

8. Examples of well-posed learning problems:
2.Sentimentanalysis:Givenasetoftextdocuments,
Task:-Istolearnamodelthatcanpredictthesentiment
ofnewdocuments(e.g.,positive,negative,orneutral).
Input:-Isthetextdata,
Output:-Isthesentimentlabel
Learningobjective:-Istominimizethepredictionerror.
PerformanceMeasure:-Percentageofpredictionof
thesentimentsofnewdocuments.
TrainingExperience:-Adatabaseofsentimentsof
givendocuments.
1.Imageclassification:Givenasetoflabeledimages,
Task:-IstolearnamodelthatcanCorrectlyclassify
newimagesintotheirrespectiveclasses.
Input:-Istheimagedata
Output:-Istheclasslabel,
Learningobjective:-IstoMinimizetheClassification
Error.
PerformanceMeasure:-Percentageofimages
correctlyclassified.
TrainingExperience:-ADatabaseofimageswith
givenclassification

9. Examples of well-posed learning problems:
3.Frauddetection:Givenasetoftransactiondata,
Task:-Istolearnamodelthatcanidentifyfraudulenttransactions.
Input:-Isthetransactiondata
Output:-Isabinarylabel(fraudulentornot),
Learningobjective:-Istominimizethefalsepositiveandfalse
negativerates.
PerformanceMeasure:-PercentageofFalsePositiveandFalse
NegativeRates.
4.Regression:Givenasetofinputfeaturesandcorrespondingtarget
values,
Task:-Taskistolearnamodelthatcanpredictthetargetvaluefor
newinputdata
Input:-Isthefeaturedata
Output:-Isthetargetvalue,
Learningobjective:-Istominimizethepredictionerror(e.g.,mean
squarederror).
PerformanceMeasure:-Percentageofthepredictionerror.

12. History of Machine Learning
Year1950:AlanTuringdevelopedtheTuringTestduringthisyear.
Year1957:Perceptron-ThefirsteverNeuralNetwork
Year1960:MITdevelopedaNaturalLanguageProcessingprogramtoactasatherapist.TheprogramwascalledELIZA.
Year1967:TheadventofNearestNeighboralgorithm,veryprominentlyusedinSearchandApproximation
Year1970:Backpropagationtakesshape.BackpropagationisasetofalgorithmsusedextensivelyinDeepLearning.
Year1980:KunihikoFukushimasuccessfullybuiltamultilayeredNeuralNetworkcalledANN.
Year1981:ExplanationBasedLearning
Year1989:ReinforcementLearningisfinallyrealized.Q-Learningalgorithm.
Year2009:ImageNet
Year2010:GoogleBrainandFacebook'sDeepFace
Year2022:ChatGPTChatGenerativePre-trainedTransformer
https://www.zeolearn.com/magazine/what-is-machine-learning

13. Artificial Intelligence vs. Machine Learning vs. Deep Learning vs. Neural
Networks
Machinelearning,Deeplearning,andNeuralnetworksareallsub-fieldsofArtificialIntelligence.
Neuralnetworksisasub-fieldofMachinelearning,andDeeplearning.
Deep"Machinelearningcanuselabeleddatasets,alsoknownasSupervisedlearning.Eliminatessomeofthe
humaninterventionrequiredandenablestheuseoflargerdatasets.
“Non-deep",Machinelearningismoredependentonhumaninterventiontolearn.Humanexpertsdetermine
thesetoffeaturestounderstandthedifferencesbetweendatainputs,requiringmorestructureddatatolearn.
Neuralnetworks,orartificialneuralnetworks(ANNs),arecomprisedofnodelayers,containinganinputlayer,
oneormorehiddenlayers,andanoutputlayer.Eachnode,orartificialneuron,connectstoanotherandhasan
associatedweightandthreshold.
DeeplearningandNeuralNetworksareaccelerateprogressinareassuchascomputervision,naturallanguage
processing,andspeechrecognition.

14. Artificial Intelligence vs. Machine Learning vs. Deep Learning vs.
Neural Networks
AIreferstothesoftwareandprocessesthataredesignedtomimicthewayhumansthinkandprocess
information.Itincludescomputervision,naturallanguageprocessing,robotics,autonomousvehicleoperating
systems,andmachinelearning.
Withthehelpofartificialintelligence,Devicesareabletolearnandidentifyinformationinordertosolve
problemsandofferkeyinsightsintovariousdomains.

15. Artificial Intelligence vs. Machine Learning vs. Deep
Learning vs. Neural Networks
AIenablesmachinestounderstanddataandmakedecisionsbasedonpatternshidden
indatawithoutanyhumanintervention.
Machinesadjusttheirknowledgebasedonnewinputs.
Example,Self-drivingcars,AlexaandCortana-Conversationswithusinournatural
humanlanguage
MachineLearning:-SubsetofAI
Machinelearningwiththehelpofthealgorithmcanprocessthesurplusof
informationandoutputanaccuratepredictionwithinmoments.Usedeeplearningall
thetime.
Usesstatisticalmodelstoexplore,analyzeandfindpatternsinlargeamountsof
data.
Performtaskswithoutbeingexplicitlyprogrammed,allowsthemtolearnfrom
experienceandimproveovertimewithouthumanintervention.
https://learnerjoy.com/artificial-intelligence-vs-machine-learning-vs-deep-learning-vs-data-science/

16. Artificial Intelligence vs. Machine Learning vs. Deep
Learning vs. Neural Networks
Approaches:-1.Supervisedlearning,2.Unsupervisedlearningand3.
Reinforcementlearning.
1.Supervisedlearning:-Requiresahumantoinputlabelleddata/Past
Labeleddataintothemachineandoutputsapredictionofanewsample.
2.Unsupervisedlearning:-Takesunlabeleddataasinput,groupsthe
databasedonitssimilarityandoutputsclustersofsimilarsamplesforthe
humantoanalyzefurtherreinforcement.O/pNotknown.Algorithms-L-
means,HierarchicalClustering,PCA,NeuralNetwork.
3.Reinforcementlearning.:-Reinforcementlearningisalsoknownas
semi-supervisedlearning.Asmallamountoflabeleddataandalarge
amountofunlabeleddataandutilizesarewardortrialanderrorsystem
tolearnovertime.GoodActionandBadAction

17. Artificial Intelligence vs. Machine Learning vs. Deep
Learning vs. Neural Networks
DeepLearning-Deeplearningisthesubsetofmachinelearning.
Themainideabehinddeeplearningismachinestolearnthingslikethehuman
brain.
Humanbrainismadeofmultitudesofneuronsthatallowustooperatetheway
wedo.
Thecollectionofconnectedneuronsinahumanbrain,scientistscreateamulti-
layernetworkthatmachinescouldusetolearnfromexperienceandpredict.
Techniques
ArtificialNeuralNetworks(ANN):-I/PintheformofNumbers
ConvolutionalNeuralNetworks(CNN):-I/PintheformofImages
Recurrentneuralnetworks(RNN).I/PintheformofTimeSeriesData
Two popular frameworks used in Deep learning are
•PyTorch by Facebook
•TensorFlow by Google

18. Artificial Intelligence vs. Machine Learning vs. Deep
Learning vs. Neural Networks
DataScience
Datascienceistoperformexploratoryanalysistobetterunderstand
thedata.
ItplaysahugerolewhenbuildingMLmodels.Ifyouhaveahuge
amountofdata,youwillgetmoreinsightsfromdataandaccurate
resultsthatcanbeappliedtobusinessusecases.
Statisticaltools–Linearalgebra

19. 

20. 

21. 

22. 

23. 

24. 

25. 

26. 

27. 

28. 

29. 

30. 

31. 

32. 

33. 

34. 

35. 

36. 

37. 

38. 

39. 

40. 

41. 

42. 

43. 

44. 

45. 

46. 

47. 

48. 

49. 

50. 

51. 

52. 

53. 

54. 

55. 

56. 

57. 

58. 

59. 

60. 

61. 

62. 

63. 

64. 

65. 

66. 

67. 

68. 

69. 

70. 

71. 

72. 

73. 

74. 

75. 

76. 

77. 

78. 