Machine Learning Introduction. From basic to Deep Learning.

Machine Learning Introduction from Basic to Deep Learning

Applied Areas • Image Recognition, Voice Recognition • Text Translation
• Ad Tech • Medical • Biology • Industry

Linear Regression

Prediction of Housing Price BSFB N QSJDF 64% )PXNVDIJTJUXIFO UIFBSFBJTN

Collect Data BSFB N QSJDF 64% 0VUQVU -BCFM XIBUXFXBOUUPLOPX *OQVU
'FBUVSF UPQSFEJDUPVUQVU

Cost BSFB N QSJDF 64% $PTU .PEFMZВ В Y

Cost Function: J(θ0 , θ1 ) Find θ1 and θ2
which minimizes J(θ0 , θ1 ). How to get an optimum model

Predict BSFB N QSJDF 64% .PEFMZВ В Y )PXNVDIJTJUXIFOUIF BSFBJTN

Neural Network

Categorical Classiﬁcation +BWB ZFBST ZFBST ZFBST ZFBST 3VCZ ZFBST ZFBST
ZFBST ZFBST 0CK$ ZFBST ZFBST ZFBST ZFBST 1)1 ZFBST ZFBST ZFBST ZFBST "OESPJE %FWFMPQFS J04 %FWFMPQFS 8FC %FWFMPQFS -BCFM 'FBUVSFT &YQFSJFODF

Neural Network +BWB
3VCZ 0CK$ 1)1 "OESPJE J04 8FC

Neural Network Y Y Y Y BZ BZ BZ +BWB
3VCZ 0CK$ 1)1 "OESPJE J04 8FC В В В В В В В В В В В В ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y *PNJUUFECJBTUFSN I B

Neural Network Y Y Y Y BZ BZ BZ +BWB
3VCZ 0CK$ 1)1 "OESPJE J04 8FC *PNJUUFECJBTUFSN В В В I B

Neural Network ZI В Y В Y В Y В
Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y *PNJUUFECJBTUFSN N N N ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y ZI В Y В Y В Y В Y

Neural Network :I 9Ϻ В В В В В В
В В В В В В Y Y Y Y Y Y Y Y Y Y Y Y Z Z Z Z Z Z Z Z Z Y NYO Y Y В В В В В В В В В В В В

Calculate an example of Y
&YBNQMFPG5IFUBϺFY 0VUQVU:FY I B 9

Cost
:FY : Cost : J(Θex ) = (1-0.18)2 + (0.55-0)2 + (0.26-0)2 + (0.38-0)2 + (0.40-1)2 + (0.21-0)2 + (0.40-0)2 + (0.23-0)2 + (0.36-1)2

How to get an optimum model Cost Function: J(Θ) Find
Θ which minimizes J(Θ).

More complex model Y Y Y Y BG B BZ
BZ +BWB 3VCZ 0CK$ 1)1 "OESPJE J04 8FC *PNJUUFECJBTUFSN G B JTBOPOMJOFBSGVODUJPO BZ )JEEFO-BZFS I B Ϻ BG B BG B B G B B G B Ϻ

Model Evaluation

Low cost always does not mean good model.

Overﬁtting BSFB N QSJDF 64% 5IPVHIUIFDPTUJTWFSZMPX JTUIJTBHPPENPEFM

Test Data BSFB N QSJDF 64% (JWFZPVSNPEFM5FTU%BUBXIJDIXFSFOPUVTFEJOUSBJOJOH &WBMVBUFUIFDPTUBHBJOTUUIF5FTU%BUB

Hyper Parameter Tuning w 4IPVME*BEENPSFVOJUT w 4IPVME*BEENPSFIJEEFOMBZFST w
*OJUJBMWBMVFTGPSUIFUB w -FBSOJOHSBUF FUD 9 : 9 : PS

Cross Validation Data 8IFOXFBSFUVOJOHIZQFSQBSBNFUFST XFTIPVMEOPUFWBMVBUFUIFNCZ5FTU%BUB 0UIFSXJTF UIFIZQFSQBSBNFUFSTXJMMCFPQUJNJ[FEGPS 5FTU%BUB5IFO ZPVSNPEFMXJMMMPTFHFOFSBMJ[BUJPO
4QMJU5SBJOJOH%BUBBOEVTFJUBT$SPTT7BMJEBUJPO %BUB XIJDIFWBMVBUFIZQFSQBSBNFUFST 5IF5FTU%BUBNVTUCFVTFEPOMZGPSpOBMFWBMVBUJPO

Data splitting $PMMFDU%BUB 4QMJU%BUB 5SBJOBOE$SPTT 7BMJEBUFNPEFM &WBMVBUFNPEFM 5SBJOJOH%BUB $SPTT7BMJEBUJPO%BUB 5FTU%BUB

Further studying • How to search optimum (Gradient descent) •
How to prevent overﬁtting (Regularization) • How to process images effectively (CNN) • How to process sequential data (RNN) • Other methods except Neural Network

Demonstration

Machine Learning Introduction. From basic to Deep Learning.

Machine Learning Introduction. From basic to Deep Learning.

akirasosa

More Decks by akirasosa

Other Decks in Technology

Featured

Transcript