1. 1 Tournament Not complete Processing will begin again tonight, 7:30PM until wee hours Friday, 8-5. Extra Credit 5 points for passing screening, in tournament 5 points for top 60% (top 16) 10 points for top 8 10 points for top 4 10 points for top 1 Celebration next Tuesday: it’s a party! Be prepared to talk about the evaluation function you used What worked Next class: tutorial on machine learning tools

2. Medical Application of Bayesian Networks: Pathfinder

3. 3 Pathfinder Domain: hematopathology diagnosis Microscopic interpretation of lymph-node biopsies Given: 100s of histologic features appearing in lymph node sections Goal: identify disease type malignant or benign Difficult for physicians

4. 4 Pathfinder System Bayesian Net implementation Reasons about 60 malignant and benign diseases of the lymph node Considers evidence about status of up to 100 morphological features presenting in lymph node tissue Contains 105,000 subjectively-derived probabilities

5. 5

6. 6 Commercialization Intellipath Integrates with videodisc libraries of histopathology slides Pathologists working with the system make significantly more correct diagnoses than those working without Several hundred commercial systems in place worldwide

7. 7 Sequential Diagnosis

8. 8 Features Structured into a set of 2-10 mutually exclusive values Pseudofollicularity Absent, slight, moderate, prominent Represent evidence provided by a feature as F 1,F 2, … F n

9. 9 Value of information User enters findings from microscopic analysis of tissue Probabilistic reasoner assigns level of belief to different diagnoses Value of information determines which tests to perform next Full disease utility model making use of life and death decision making Cost of tests Cost of misdiagnoses

10. 10

11. 11

12. 12 Group Discrimination Strategy Select questions based on their ability to discriminate between disease classes For given differential diagnoisis, select most specific level of hierarchy and selects questions to discriminate among groups Less efficient Larger number of questions asked

13. 13

14. 14

15. 15 Other Bayesian Net Applications Lumiere – Who knows what it is?

16. 16 Other Bayesian Net Applications Lumiere Single most widely distributed application of BN Microsoft Office Assistant Infer a user’s goals and needs using evidence about user background, actions and queries VISTA Help NASA engineers in round-the-clock monitoring of each of the Space Shuttle’s orbiters subsystem Time critical, high impact Interpret telemetry and provide advice about likely failures Direct engineers to the best information In use for several years Microsoft Pregnancy and Child Care What questions to ask next to diagnose illness of a child

17. Machine Learning Reading: Chapter 18

18. 18 Machine Learning and AI Improve task performance through observation, teaching Acquire knowledge automatically for use in a task Learning as a key component in intelligence

19. 19 Inductive Learning Input: x, f(x) Output: a function h that approximates f A good hypothesis, h, is a generalization or learned rule

20. 20 How do systems learn? Supervised Unsupervised Reinforcement

21. 21 Three Types of Learning Rule induction E.g., decision trees Knowledge based E.g., using a domain theory Statistical E.g., Naïve bayes, Nearest neighbor, support vector machines

22. 22 Applications Language/speech Machine translation Summarization Grammars IR Text categorization, relevance feedback Medical Assessment of illness severity Vision Face recognition, digit recognition, outdoor scene recognition Security Intrusion detection, network traffic, credit fraud Social networks Email traffic To think about: applications to systems, computer engineering, software?

23. 23 Language Tasks Text summarization Task: given a document which sentences could serve as the summary Training data: summary + document pairs Output: rules which extract sentences given an unseen document Grammar induction Task: produce a tree representing syntactic structure given a sentence Training data: set of sentences annotated with parse tree Output: rules which generate a parse tree given an unseen sentence

24. 24 IR Task Text categorization http://www.yahoo.comwww.yahoo.com Task: given a web page, is it news or not? Binary classification (yes, no) Classify as one of business&economy,news&media, computer Training data: documents labeled with category Output: a yes/no response for a new document; a category for a new document

25. 25 Medical Task: Does a patient have heart disease (on a scale from 1 to 4) Training data: Age, sex,cholesterol, chest pain location, chest pain type, resting blood pressure, smoker?, fasting blood sugar, etc. Characterization of heart disease (0,1-4) Output: Given a new patient, classification by disease

26. 26 General Approach Formulate task Prior model (parameters, structure) Obtain data What representation should be used? (attribute/value pairs) Annotate data Learn/refine model with data (training) Use model for classification or prediction on unseen data (testing) Measure accuracy

27. 27 Issues Representation How to map from a representation in the domain to a representation used for learning? Training data How can training data be acquired? Amount of training data How well does the algorithm do as we vary the amount of data? Which attributes influence learning most? Does the learning algorithm provide insight into the generalizations made?

28. 28 Classification Learning Input: a set of attributes and values Output: discrete valued function Learning a continuous valued function is called regression Binary or boolean classification: category is either true or false

29. 29 Learning Decision Trees Each node tests the value of an input attribute Branches from the node correspond to possible values of the attribute Leaf nodes supply the values to be returned if that leaf is reached

30. 30 Example http://www.ics.uci.edu/~mlearn/MLSummary.html Iris Plant Database Which of 3 classes is a given Iris plant? Iris Setosa Iris Versicolour Iris Virginica Attributes Sepal length in cm Sepal width in cm Petal length in cm Petal width in cm

31. 31 Summary Statistics: Min Max Mean SD ClassCorrelation sepal length: 4.3 7.9 5.84 0.83 0.7826 sepal width: 2.0 4.4 3.05 0.43 -0.4194 petal length: 1.0 6.9 3.76 1.76 0.9490 (high!) petal width: 0.1 2.5 1.20 0.76 0.9565 (high!) Rules to learn If sepal length > 6 and sepal width > 3.8 and petal length < 2.5 and petal width < 1.5 then class = Iris Setosa If sepal length > 5 and sepal width > 3 and petal length >5.5 and petal width >2 then class = Iris Versicolour If sepal length 3 and petal length  2.5 and ≤ 5.5 and petal width  1.5 and ≤ 2 then class = Iris Virginica

32. 32 Data S-lengthS-widthP-lengthP-widthClass 16.836.32.3Versicolour 273.92.41.1Setosa 3232.61.7Verginica 433.42.51.5Verginica 55.53.66.82.4Versicolour 67.74.11.21.4Setosa 76.34.31.61.2Setosa 813.72.82Verginica 964.25.62.1Versicolour

33. 33 Data S-lengthS-widthP-lengthClass 16.836.3Versicolour 273.92.4Setosa 3232.6Verginica 433.42.5Verginica 55.53.66.8Versicolour 67.74.11.2Setosa 76.34.31.6Setosa 813.72.8Verginica 964.25.6Versicolour

34. 34 Constructing the Decision Tree Goal: Find the smallest decision tree consistent with the examples Find the attribute that best splits examples Form tree with root = best attribute For each value v i (or range) of best attribute Selects those examples with best=v i Construct subtree i by recursively calling decision tree with subset of examples, all attributes except best Add a branch to tree with label=v i and subtree=subtree i

35. 35 Construct example decision tree

36. 36 Issues Representation How to map from a representation in the domain to a representation used for learning? Training data How can training data be acquired? Amount of training data How well does the algorithm do as we vary the amount of data? Which attributes influence learning most? Does the learning algorithm provide insight into the generalizations made?