Outline Wh y Mac hine Learning� � What is a w ell�de�ned learning problem� � An example� learning to pla y c hec k ers � What questions should w e ask ab out Mac hine � Learning� � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Wh y Mac hine Learning Recen t progress in algorithms and theory � Gro wing �o o d of online data � Computational p o w er is a v ailable � Budding industry � Three nic hes for mac hine learning� � Data mining � using historical data to impro v e decisions medical records medical kno wledge � � Soft w are applications w e can�t program b y hand � autonomous driving � sp eec h recognition � Self customizing programs � Newsreader that learns user in terests � � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
T ypical Datamining T ask Data� ... Patient103 Patient103 Patient103 time=n time=1 time=2 Age: 23 Age: 23 Age: 23 FirstPregnancy: no FirstPregnancy: no FirstPregnancy: no Anemia: no Anemia: no Anemia: no Diabetes: no Diabetes: no Diabetes: YES PreviousPrematureBirth: no PreviousPrematureBirth: no PreviousPrematureBirth: no Ultrasound: ? Ultrasound: abnormal Ultrasound: ? Elective C−Section: ? Elective C−Section: no Elective C−Section: no Emergency C−Section: ? Emergency C−Section: ? Emergency C−Section: Yes ... ... ... Giv en� ���� patien t records� eac h describing a � pregnancy and birth Eac h patien t record con tains ��� features � Learn to predict� Classes of future patien ts at high risk for � Emergency Cesarean Section � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Datamining Result Data� ... Patient103 Patient103 Patient103 time=n time=1 time=2 Age: 23 Age: 23 Age: 23 FirstPregnancy: no FirstPregnancy: no FirstPregnancy: no Anemia: no Anemia: no Anemia: no Diabetes: no Diabetes: no Diabetes: YES PreviousPrematureBirth: no PreviousPrematureBirth: no PreviousPrematureBirth: no Ultrasound: ? Ultrasound: abnormal Ultrasound: ? Elective C−Section: ? Elective C−Section: no Elective C−Section: no Emergency C−Section: ? Emergency C−Section: ? Emergency C−Section: Yes ... ... ... One of �� learned rules� If No previous vaginal delivery� and Abnormal �nd Trimester Ultrasound� and Malpresentation at admission Then Probability of Emergency C�Section is ��� Over training data� ����� � ���� Over test data� ����� � ��� � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Credit Risk Analysis Data� ... Customer103: (time=t0) Customer103: (time=t1) Customer103: (time=tn) Years of credit: 9 Years of credit: 9 Years of credit: 9 Loan balance: $2,400 Loan balance: $3,250 Loan balance: $4,500 Income: $52k Income: ? Income: ? Own House: Yes Own House: Yes Own House: Yes Other delinquent accts: 2 Other delinquent accts: 2 Other delinquent accts: 3 Max billing cycles late: 3 Max billing cycles late: 4 Max billing cycles late: 6 Profitable customer?: No Profitable customer?: ? Profitable customer?: ? ... ... ... Rules learned from syn thesized data� If Other�Delinquent�A ccoun ts � �� and Number�Delinquent� Billi ng�Cy cles � � Then Profitable�Custome r� � No �Deny Credit Card application� If Other�Delinquent�A ccoun ts � �� and �Income � ���k� OR �Years�of�Credit � �� Then Profitable�Custome r� � Yes �Accept Credit Card application� � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Other Prediction Problems Customer purc hase b eha vior� ... Customer103: (time=t0) Customer103: (time=t1) Customer103: (time=tn) Sex: M Sex: M Sex: M Age: 53 Age: 53 Age: 53 Income: $50k Income: $50k Income: $50k Own House: Yes Own House: Yes Own House: Yes MS Products: Word MS Products: Word MS Products: Word Computer: 386 PC Computer: Pentium Computer: Pentium Purchase Excel?: ? Purchase Excel?: ? Purchase Excel?: Yes ... ... ... Customer reten tion� ... Customer103: (time=t0) Customer103: (time=t1) Customer103: (time=tn) Sex: M Sex: M Sex: M Age: 53 Age: 53 Age: 53 Income: $50k Income: $50k Income: $50k Own House: Yes Own House: Yes Own House: Yes Checking: $5k Checking: $20k Checking: $0 Savings: $15k Savings: $0 Savings: $0 Current−customer?: No Current−customer?: yes ... Current−customer?: yes ... Pro cess optimization� ... Product72: (time=t0) Product72: (time=t1) Product72: (time=tn) Stage: mix Stage: cook Stage: cool Mixing−speed: 60rpm Temperature: 325 Fan−speed: medium Viscosity: 1.3 Viscosity: 3.2 Viscosity: 1.3 Fat content: 15% Fat content: 12% Fat content: 12% Density: 2.8 Density: 1.1 Density: 1.2 Spectral peak: 2800 Spectral peak: 3200 Spectral peak: 3100 Product underweight?: Yes Product underweight?: ?? Product underweight?: ?? ... ... ... � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Problems T o o Di�cult to Program b y Hand AL VINN �P omerleau� driv es �� mph on high w a ys Sharp Straight Sharp Left Ahead Right 30 Output Units 4 Hidden Units 30x32 Sensor Input Retina � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Soft w are that Customizes to User h ttp���www�wisewi re�com � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Where Is this Headed� T o da y� tip of the iceb erg First�generation algorithms� neural nets� � decision trees� regression ��� � Applied to w ell�formated database Budding industry � Opp ortunit y for tomorro w� enormous impact Learn across full mixed�media data � Learn across m ultiple in ternal databases� plus � the w eb and newsfeeds Learn b y activ e exp erimen tation � Learn decisions rather than predictions � Cum ulativ e� lifel ong learning � Programm ing languages with learning � em b edded� � lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Relev an t Discipli nes Arti�cial in telli gence � Ba y esian metho ds � Computational complexit y theory � Con trol theory � Information theory � � Philosoph y Psyc hology and neurobiology � Statistics � � � � � �� lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
What is the Learning Problem� Learning � Impro ving with exp erience at some task Impro v e o v er task � � T with resp ect to p erformance measure � � P based on exp erience � � E E�g�� Learn to pla y c hec k ers � Pla y c hec k ers � T � � of games w on in w orld tournamen t � P � opp ortunit y to pla y against self � E �� lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Learning to Pla y Chec k ers � Pla y c hec k ers � T � P ercen t of games w on in w orld tournamen t � P What exp erience� � What exactly should b e learned� � Ho w shall it b e represen ted� � What sp eci�c algorithm to learn it� � �� lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
T yp e of T raining Exp erience Direct or indirect� � T eac her or not� � A problem� is training exp erience represen tativ e of p erformance goal� �� lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Cho ose the T arget F unction � �� � C hooseM ov e B oar d � M ov e � �� � V B oar d � � ��� � �� lecture slides for textb o ok Machine L e arning � T� Mitc hell� McGra w Hill� ����
Recommend
More recommend
