Machine Learning Machine Learning Tom T Mitchell Mc
- Slides: 44
Machine Learning
교재 § Machine Learning, Tom T. Mitchell, Mc. Graw-Hill 일부 § Reinforcement Learning: An Introduction, R. S. Sutton and A. G. Barto, The MIT Press, 1998 발표 2
Machine Learning § How to construct computer programs that automatically improve with experience § Data mining(medical applications: 1989), fraudulent credit card (1989), transactions, information filtering, users’ reading preference, autonomous vehicles, backgammon at level of world champions(1992), speech recognition(1989), optimizing energy cost § Machine learning theory – How does learning performance vary with the number of training examples presented – What learning algorithms are most appropriate for various types of learning tasks 3
예제 프로그램 § http: //www. cs. cmu. edu/~tom/mlbook. html – Face recognition – Decision tree learning code – Data for financial loan analysis – Bayes classifier code – Data for analyzing text documents 4
이론적 연구 § Fundamental relationship among the number of training examples observed, the number of hypotheses under consideration, and the expected error in learned hypotheses § Biological systems 5
Def. A computer program is said to learn from experience E wrt some classes of tasks T and performance P, if its performance at tasks in T, as measured by P, improves with experience E. 6
Outline § § Why Machine Learning? What is a well-defined learning problem? An example: learning to play checkers What questions should we ask about Machine Learning? 7
Why Machine Learning § § Recent progress in algorithms and theory Growing flood of online data Computational power is available Budding industry 8
Three niches for machine learning: § Data mining : using historical data to improve decisions – medical records medical knowledge § Software applications we can't program by hand – autonomous driving – speech recognition § Self customizing programs – Newsreader that learns user interests 9
Typical Datamining Task (1/2) § Data : 10
Typical Datamining Task (2/2) § Given: – 9714 patient records, each describing a pregnancy and birth – Each patient record contains 215 features § Learn to predict: – Classes of future patients at high risk for Emergency Cesarean Section 11
Datamining Result One of 18 learned rules: If No previous vaginal delivery, and Abnormal 2 nd Trimester Ultrasound, and Malpresentation at admission Then Probability of Emergency C-Section is 0. 6 Over training data: 26/41 =. 63, Over test data: 12/20 =. 60 12
Credit Risk Analysis (1/2) § Data : 13
Credit Risk Analysis (2/2) Rules learned from synthesized data: If Other-Delinquent-Accounts > 2, and Number-Delinquent-Billing-Cycles > 1 Then Profitable-Customer? = No [Deny Credit Card application] If Other-Delinquent-Accounts = 0, and (Income > $30 k) OR (Years-of-Credit > 3) Then Profitable-Customer? = Yes [Accept Credit Card application] 14
Other Prediction Problems (1/2) 15
Other Prediction Problems (2/2) 16
Problems Too Difficult to Program by Hand § ALVINN [Pomerleau] drives 70 mph on highways 17
Software that Customizes to User http: //www. wisewire. com 18
Where Is this Headed? (1/2) § Today: tip of the iceberg – First-generation algorithms: neural nets, decision trees, regression. . . – Applied to well-formatted database – Budding industry 19
Where Is this Headed? (2/2) § Opportunity for tomorrow: enormous impact – Learn across full mixed-media data – Learn across multiple internal databases, plus the web and newsfeeds – Learn by active experimentation – Learn decisions rather than predictions – Cumulative, lifelong learning – Programming languages with learning embedded? 20
빅 데이터를 활용한 분석 영역은 무한합니다. Smarter Healthcare Multi-channel sales Finance Log Analysis Homeland Security Traffic Control Telecom Search Quality Fraud and Risk Retail: Churn, NBO Manufacturing Trading Analytics
Hadoop 활용 사례 : Yahoo & Visa • Hadoop at Yahoo! ü 25, 000+ machines in 10+ clusters (largest is 3, 000 machines) ü 3 PBs of data (compressed, unreplicated) ü 10, 000+ jobs/week • Hadoop@Visa ü 2년치 raw transaction data를 이용하여 real-time risk scoring system에 사용될 데이타 요소들을 생성 ü 500 M distinct accounts, 100 M transactions per day, 200 bytes per transaction, 2 years total 73 B transactions (36 TB) ü Processing time : 1 months 13 minutes (3000 times faster)
Relevant Disciplines § § § § § Artificial intelligence Bayesian methods Computational complexity theory Control theory Information theory Philosophy Psychology and neurobiology Statistics. . . 31
What is the Learning Problem? § Learning = Improving with experience at some task – Improve over task T, – with respect to performance measure P, – based on experience E. § E. g. , Learn to play checkers – T: Play checkers – P: % of games won in world tournament – E: opportunity to play against self 32
Learning to Play Checkers § § § T: Play checkers P: Percent of games won in world tournament What experience? What exactly should be learned? How shall it be represented? What specific algorithm to learn it? 33
Type of Training Experience § Direct or indirect? § Teacher or not? A problem: is training experience representative of performance goal? 34
Choose the Target Function § Choose. Move : Board Move ? ? § V : Board R ? ? §. . . 35
Possible Definition for Target Function V § § if b is a final board state that is won, then V(b) = 100 if b is a final board state that is lost, then V(b) = -100 if b is a final board state that is drawn, then V(b) = 0 if b is not a final state in the game, then V(b) = V(b'), where b' is the best final board state that can be achieved starting from b and playing optimally until the end of the game. This gives correct values, but is not operational 36
Choose Representation for Target Function § § collection of rules? neural network ? polynomial function of board features? . . . 37
A Representation for Learned Function w 0+ w 1·bp(b)+w 2·rp(b)+w 3·bk(b)+w 4·rk(b)+w 5·bt(b)+w 6·rt(b) § § § bp(b) : number of black pieces on board b rp(b) : number of red pieces on b bk(b) : number of black kings on b rk(b) : number of red kings on b bt(b) : number of red pieces threatened by black (i. e. , which can be taken on black's next turn) § rt(b) : number of black pieces threatened by red 38
Obtaining Training Examples § § § V(b): the true target function ^ V(b) : the learned function Vtrain(b): the training value One rule for estimating training values: ^ § Vtrain(b) V(Successor(b)) 39
Choose Weight Tuning Rule LMS Weight update rule: Do repeatedly: § Select a training example b at random 1. Compute error(b): error(b) = Vtrain(b) – V(b) 2. For each board feature fi, update weight wi: wi + c · fi · error(b) c is some small constant, say 0. 1, to moderate the rate of learning 40
Final design § The performance system – Playing games § The critic – 차이 발견 (분석) § The generalizer – Generate new hypothesis § The experiment generator – Generate new problems 41
Design Choices 43
Some Issues in Machine Learning § What algorithms can approximate functions well (and when)? § How does number of training examples influence accuracy? § How does complexity of hypothesis representation impact it? § How does noisy data influence accuracy? § What are theoretical limits of learnability? § How can prior knowledge of learner help? § What clues can we get from biological learning systems? § How can systems alter their own representations? 44
- Cmu tom mitchell
- Machine learning tom
- Pac learning model in machine learning
- Tom tom go 910
- Symbols in the devil and tom walker
- Concept learning task in machine learning
- Analytical learning in machine learning
- Pac learning model in machine learning
- Inductive and analytical learning
- Inductive analytical approach to learning
- Instance based learning in machine learning
- Inductive learning machine learning
- First order rule learning in machine learning
- Eager learning
- Cmu machine learning
- Turbine banki-mitchell ou crossflow
- Nickole mitchell
- Antony flew theology and falsification
- Mitchell1 social crm reviews
- Retinopatia hipertensiva kwb
- Clasificacion de wong mitchell
- Classification de mitchell ostéonécrose
- Mitchell daysh
- 4 table mitchell movement
- Mitchell on demand pro
- Michael c mitchell
- Christina tripodi mitchell
- Mitchell merritt algorithm
- Dr mitchell charap
- Emmett mitchell
- Mitchell and cheeseman
- Mitchell cairo md
- Mitchell sanders the things they carried
- Stephen mitchell md
- Dr charlotte mitchell
- Susan o shea
- Mitchell lipton md
- John mitchell stanford
- Nickole mitchell
- Silas weir mitchell back injury
- Estilo de vida vals
- Caroline mitchell mgh
- World power era eoc blitz review
- Dr ralph mitchell
- Mitchell feldman md