SlideShare una empresa de Scribd logo
1 de 10
University of Wisconsin – Madison
                                 Computer Sciences Department



                     CS 760 - Machine Learning
                                             Spring 2003

                                                Exam
                                      7:15-9:15pm, May 6, 2003
                                     Room 3345 Engineering Hall

                                            CLOSED BOOK
                              (one sheet of notes and a calculator allowed)


Write your answers on these pages and show your work. If you feel that a question is not fully
specified, state any assumptions that you need to make in order to solve the problem. You may
use the backs of these sheets for scratch work.

Write your name on this and all other pages of this exam. Make sure your exam contains
5 problems on 10 pages.


       Name           ________________________________________________________________


       Student ID     ________________________________________________________________


                      Problem                   Score           Max Score


                          1                     ______               40

                          2                     ______               17

                          3                     ______               10

                          4                     ______               12

                          5                     ______               21


                      TOTAL                     ______              100
Name: _______________________________________


Problem 1 – Learning from Labeled Examples (40 points)

Imagine that you are given the following set of training examples.
Feature F1 can take on the values a, b, or c; Feature F2 is Boolean-valued;
and Feature F3 is always a real-valued number in [0,1].

                               F1    F2        F3       Category

               Example 1       a      T        0.2         +
               Example 2       b      F        0.5         +
               Example 3       b      F        0.9         +
               Example 4       b      T        0.6         –
               Example 5       a      F        0.1         –
               Example 6       a      T        0.7         –

a) How might a Naive Bayes system classify the following test example?
   Be sure to show your work. (Discretize the numeric feature into three equal-width bins.)

               F1 = c      F2 = T   F3 = 0.8




b) Describe how a 2-nearest-neighbor algorithm might classify Part a’s test example.




                                           Page 2 of 10
Name: _______________________________________




c) Show the calculations that ID3 would perform to determine the root node of a decision tree
     using the above training examples.




d) Using the “decision stump” learner as boosting’s weak learner, what would the weights be on
     each of the six training examples after one (1) round of AdaBoost? Show your work.

        Weight on Example 1:   _________
        Weight on Example 2:   _________
        Weight on Example 3:   _________
        Weight on Example 4:   _________
        Weight on Example 5:   _________
        Weight on Example 6:   _________




e)   Show the feature space for this training set. Use only features F2 and F3 (ie, for this part,
     ignore F1). Show what a linear support-vector machine might learn from this training set.
     Briefly explain your answer.




                                             Page 3 of 10
Name: _______________________________________




f) Create a neural network with no hidden units and a sigmoidal output unit. Initialize all the
   free parameters to 2 and use a learning rate of 0.1. Use squared error is the error function.

       i. Draw a picture of this neural network before training.




       ii. Calculate ∂ Error / ∂ weighti for this network. Recall that the derivative of the
         sigmoid is output × (1 – output).




       iii. Show how the free parameters would change after backpropagating on the first
         training example in the table above. Indicate the weight changes on the neural network
         you drew above and show your work below.




       iv. According to Rumelhart, which error function should one use on classification tasks
         such as this one? And how should one interpret the neural network’s output?




                                             Page 4 of 10
Name: _______________________________________




Problem 2 – Reinforcement Learning (17 points)

Consider the deterministic reinforcement environment drawn below. The numbers on the arcs
indicate the immediate rewards. Let the discount rate equal 0.5.


                                 r = -10                    -15
                     sta                           a                          b
                     rt                                            15
                                                            6
                                 10                    -6         -3      3


                                             -2                   c 5
                     en
                     d
                            0



Assume that the current Q table is represented by the Q values on the arcs on the environment's
state-action graph below. Also assume that the learner is currently in Node b.



                                 Q = -1                     -7
                     sta                           a                          b
                     rt                                            14
                                                            7
                                  12                   -1         -9     1


                                              -3                  c 5
                      en
                      d
                             0


a) The learner chooses to do an exploration step out of Node b, using the “softmax” function.
   What is the probability is that the learner will choose the action that leads to Node a (be sure
   to state any assumptions you need to make in order to compute this probability)?




                                            Page 5 of 10
Name: _______________________________________




b) Assume that the learner is magically transported to Node c and that the Q table is still in the
   state shown below. For this question, assume the learner chooses to first go to Node b and
   then to Node a. Be sure to show the calculations that produce your answers to the questions
   below.
                                Q = -1                     -7
                     sta                          a                         b
                     rt                                          14
                                                           7
                                12                    -1        -9      1


                                             -3                 c 5
                     en
                     d
                            0


      i. Using one-step Q learning, what new value would there be on the c→ b arc?




     ii. Using one-step SARSA learning, what new value would there be on the c→ b arc?




c) This time assume the learner is currently in Node start and the Q table contains the values in
   the graph above (the values before you answered Part b). The learner does three (3)
   exploitation steps and employs two-step Q learning along the way. Indicate on the graph
   above what changes in the Q table this time. Show your calculations below.




                                           Page 6 of 10
Name: _______________________________________


Problem 3 – Computational Learning Theory (10 points)

a) Assume that we have the predictions below of five experts, as well as the correct answer.
   Using the weighted majority algorithm (with β=0.5) in order to track the best expert, show
   how the weight given to each expert changes after each example. Show your work.

       Expert 1         2        3        4        5                 Correct Answer

               T        T        T        F        F                        F
       Weights ________________________________________




               F        T        F        T        T                        T
       Weights ________________________________________




               T        F        F        F        T                        F
       Weights ________________________________________




b) Consider the Boolean concept class of circles in the space of two numeric features (points on
   or inside the circle are considered positive). What is the VC-dimension for this concept
   space? Explain.




                                                  Page 7 of 10
Name: _______________________________________


Problem 4 – Overfitting Avoidance (12 points)
a) Use a Venn diagram of feature space to illustrate how a decision tree might overfit a training
   set. Briefly explain your answer.




b) Briefly describe how one can choose the number of hidden units to use in a neural network in
   order to reduce the chances of overfitting.




c) Informally (i.e., verbally or pictorially) explain how using a kernel might lead to overfitting
   in a support-vector machine.




d) Outline a methodology where an experimenter incorrectly uses cross validation to set
   parameters and estimate future accuracy.




                                            Page 8 of 10
Name: _______________________________________


Problem 5 – Short Discussion Questions (21 points)

a)   Using a Venn diagram, qualitatively illustrate how 1-NN (one nearest-neighbor) partitions
     feature space. Assume a Boolean classification task. Briefly explain your answer.




b) Consider creating an ensemble of eleven support-vector machines (SVM’s) using bagging
     and a standard linear SVM. Dataset A has 100 support vectors and Dataset B has 1000
     support vectors (when run on all the data); both datasets contain 10,000 examples and on
     both datasets the 10-fold cross-validation accuracy of a single SVM is 80%. On which
     dataset would you expect bagging to improve accuracy the most? Why?




c) What does transduction mean in machine learning?




                                            Page 9 of 10
Name: _______________________________________


d) Explain the role of slack variables in support vector machines.




e)   What do you feel is the most important improvement that PAC learning offers over learning
     in the limit? Explain your answer.




f)   Noah Taul does a proper 10-fold cross-validation of his new algorithm compared to
     backprop. A paired t-test shows he cannot disregard the null hypothesis.
     What is the null hypothesis? And how does a t-test allow one to address this hypothesis?




                                         Have a good vacation!




                                           Page 10 of 10

Más contenido relacionado

La actualidad más candente

A fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelsA fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelseSAT Journals
 
20111101151326 final hbmt4403_jan2010_updated
20111101151326 final hbmt4403_jan2010_updated20111101151326 final hbmt4403_jan2010_updated
20111101151326 final hbmt4403_jan2010_updateddtrb
 
Alg1 power points_-_unit_7_-_exponents_and_exponential_functions
Alg1 power points_-_unit_7_-_exponents_and_exponential_functionsAlg1 power points_-_unit_7_-_exponents_and_exponential_functions
Alg1 power points_-_unit_7_-_exponents_and_exponential_functionsCuteanimeboyWithhood
 
A fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelsA fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelseSAT Publishing House
 
Gaussian Processes: Applications in Machine Learning
Gaussian Processes: Applications in Machine LearningGaussian Processes: Applications in Machine Learning
Gaussian Processes: Applications in Machine Learningbutest
 
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)RAMLAH BINTI A. RANI
 
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...Jim Jenkins
 
Lecture14 - Advanced topics in association rules
Lecture14 - Advanced topics in association rulesLecture14 - Advanced topics in association rules
Lecture14 - Advanced topics in association rulesAlbert Orriols-Puig
 
Low-rank tensor methods for stochastic forward and inverse problems
Low-rank tensor methods for stochastic forward and inverse problemsLow-rank tensor methods for stochastic forward and inverse problems
Low-rank tensor methods for stochastic forward and inverse problemsAlexander Litvinenko
 
225.doc
225.doc225.doc
225.docbutest
 
Functional Programming In Mathematica
Functional Programming In MathematicaFunctional Programming In Mathematica
Functional Programming In MathematicaHossam Karim
 
Semi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster EnsembleSemi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster EnsembleAlexander Litvinenko
 
Stable matching with incomplete information
Stable matching with incomplete informationStable matching with incomplete information
Stable matching with incomplete informationTakuya Irie
 

La actualidad más candente (20)

Maths set of exercises spring 2010
Maths set of exercises spring 2010Maths set of exercises spring 2010
Maths set of exercises spring 2010
 
January 2007
January 2007January 2007
January 2007
 
Network Analysis & Synthesis Nov 2012
Network Analysis & Synthesis Nov 2012Network Analysis & Synthesis Nov 2012
Network Analysis & Synthesis Nov 2012
 
A fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelsA fusion of soft expert set and matrix models
A fusion of soft expert set and matrix models
 
Daa q.paper
Daa q.paperDaa q.paper
Daa q.paper
 
20111101151326 final hbmt4403_jan2010_updated
20111101151326 final hbmt4403_jan2010_updated20111101151326 final hbmt4403_jan2010_updated
20111101151326 final hbmt4403_jan2010_updated
 
Alg1 power points_-_unit_7_-_exponents_and_exponential_functions
Alg1 power points_-_unit_7_-_exponents_and_exponential_functionsAlg1 power points_-_unit_7_-_exponents_and_exponential_functions
Alg1 power points_-_unit_7_-_exponents_and_exponential_functions
 
A fusion of soft expert set and matrix models
A fusion of soft expert set and matrix modelsA fusion of soft expert set and matrix models
A fusion of soft expert set and matrix models
 
2007 57-p1
2007 57-p12007 57-p1
2007 57-p1
 
Gaussian Processes: Applications in Machine Learning
Gaussian Processes: Applications in Machine LearningGaussian Processes: Applications in Machine Learning
Gaussian Processes: Applications in Machine Learning
 
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)
956 Sukatan Pelajaran Matematik Lanjutan STPM (Baharu)
 
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...
Fundamentals of Engineering Probability Visualization Techniques & MatLab Cas...
 
Lecture14 - Advanced topics in association rules
Lecture14 - Advanced topics in association rulesLecture14 - Advanced topics in association rules
Lecture14 - Advanced topics in association rules
 
Low-rank tensor methods for stochastic forward and inverse problems
Low-rank tensor methods for stochastic forward and inverse problemsLow-rank tensor methods for stochastic forward and inverse problems
Low-rank tensor methods for stochastic forward and inverse problems
 
954 sp math(t)b
954 sp math(t)b954 sp math(t)b
954 sp math(t)b
 
225.doc
225.doc225.doc
225.doc
 
Functional Programming In Mathematica
Functional Programming In MathematicaFunctional Programming In Mathematica
Functional Programming In Mathematica
 
statistics assignment help
statistics assignment helpstatistics assignment help
statistics assignment help
 
Semi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster EnsembleSemi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster Ensemble
 
Stable matching with incomplete information
Stable matching with incomplete informationStable matching with incomplete information
Stable matching with incomplete information
 

Destacado

Vecer Na Temu25marec2010
Vecer Na Temu25marec2010Vecer Na Temu25marec2010
Vecer Na Temu25marec2010guestc27e91
 
Desney S Tan Curriculum Vitae
Desney S Tan Curriculum VitaeDesney S Tan Curriculum Vitae
Desney S Tan Curriculum Vitaebutest
 
ph-report.doc
ph-report.docph-report.doc
ph-report.docbutest
 
SCHOOL MISSION AND STRUCTURE
SCHOOL MISSION AND STRUCTURESCHOOL MISSION AND STRUCTURE
SCHOOL MISSION AND STRUCTUREbutest
 
Il processo creativo
Il processo creativoIl processo creativo
Il processo creativoHibo
 
Micromouse
MicromouseMicromouse
Micromousebutest
 
Dragomir R
Dragomir RDragomir R
Dragomir Rbutest
 
Andrew Shitov Rakudo Jonathan
Andrew Shitov Rakudo JonathanAndrew Shitov Rakudo Jonathan
Andrew Shitov Rakudo Jonathanguest092df8
 
Plastickachirurgie
PlastickachirurgiePlastickachirurgie
Plastickachirurgieguest19dc09
 
source2
source2source2
source2butest
 
notes as .ppt
notes as .pptnotes as .ppt
notes as .pptbutest
 
Slide 1
Slide 1Slide 1
Slide 1butest
 
learningIntro.doc
learningIntro.doclearningIntro.doc
learningIntro.docbutest
 
doc - University of Idaho
doc - University of Idahodoc - University of Idaho
doc - University of Idahobutest
 
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...butest
 

Destacado (20)

Tugas Pwer Poin
Tugas Pwer PoinTugas Pwer Poin
Tugas Pwer Poin
 
Vecer Na Temu25marec2010
Vecer Na Temu25marec2010Vecer Na Temu25marec2010
Vecer Na Temu25marec2010
 
Desney S Tan Curriculum Vitae
Desney S Tan Curriculum VitaeDesney S Tan Curriculum Vitae
Desney S Tan Curriculum Vitae
 
ph-report.doc
ph-report.docph-report.doc
ph-report.doc
 
SCHOOL MISSION AND STRUCTURE
SCHOOL MISSION AND STRUCTURESCHOOL MISSION AND STRUCTURE
SCHOOL MISSION AND STRUCTURE
 
Il processo creativo
Il processo creativoIl processo creativo
Il processo creativo
 
Micromouse
MicromouseMicromouse
Micromouse
 
Dagan
DaganDagan
Dagan
 
Dragomir R
Dragomir RDragomir R
Dragomir R
 
Andrew Shitov Rakudo Jonathan
Andrew Shitov Rakudo JonathanAndrew Shitov Rakudo Jonathan
Andrew Shitov Rakudo Jonathan
 
O R G A N I S A S I
O R G A N I S A S IO R G A N I S A S I
O R G A N I S A S I
 
Plastickachirurgie
PlastickachirurgiePlastickachirurgie
Plastickachirurgie
 
source2
source2source2
source2
 
notes as .ppt
notes as .pptnotes as .ppt
notes as .ppt
 
.doc
.doc.doc
.doc
 
Slide 1
Slide 1Slide 1
Slide 1
 
learningIntro.doc
learningIntro.doclearningIntro.doc
learningIntro.doc
 
doc - University of Idaho
doc - University of Idahodoc - University of Idaho
doc - University of Idaho
 
DIPTICO DENTOSHOP MAYO - JUNIO 15
DIPTICO DENTOSHOP MAYO - JUNIO 15DIPTICO DENTOSHOP MAYO - JUNIO 15
DIPTICO DENTOSHOP MAYO - JUNIO 15
 
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...
Julie Acker, M.S.W., CMHA Lambton Julie Acker holds a Masters ...
 

Similar a Spring 2003

Problem 1 – First-Order Predicate Calculus (15 points)
Problem 1 – First-Order Predicate Calculus (15 points)Problem 1 – First-Order Predicate Calculus (15 points)
Problem 1 – First-Order Predicate Calculus (15 points)butest
 
Homework Value of InformationPlease respond to the following.docx
Homework Value of InformationPlease respond to the following.docxHomework Value of InformationPlease respond to the following.docx
Homework Value of InformationPlease respond to the following.docxadampcarr67227
 
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docx
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docxCSCI 2033 Elementary Computational Linear Algebra(Spring 20.docx
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docxmydrynan
 
Digital electronics k map comparators and their function
Digital electronics k map comparators and their functionDigital electronics k map comparators and their function
Digital electronics k map comparators and their functionkumarankit06875
 
This quiz is open book and open notes/tutorialoutlet
This quiz is open book and open notes/tutorialoutletThis quiz is open book and open notes/tutorialoutlet
This quiz is open book and open notes/tutorialoutletBeardmore
 
Matlab 1
Matlab 1Matlab 1
Matlab 1asguna
 
Quantum Computing Notes Ver 1.2
Quantum Computing Notes Ver 1.2Quantum Computing Notes Ver 1.2
Quantum Computing Notes Ver 1.2Vijayananda Mohire
 
IRJET- Solving Quadratic Equations using C++ Application Program
IRJET-  	  Solving Quadratic Equations using C++ Application ProgramIRJET-  	  Solving Quadratic Equations using C++ Application Program
IRJET- Solving Quadratic Equations using C++ Application ProgramIRJET Journal
 
College Algebra MATH 107 Spring, 2016, V4.7 Page 1 of .docx
College Algebra   MATH 107   Spring, 2016, V4.7 Page 1 of .docxCollege Algebra   MATH 107   Spring, 2016, V4.7 Page 1 of .docx
College Algebra MATH 107 Spring, 2016, V4.7 Page 1 of .docxclarebernice
 
Huong dan cu the svm
Huong dan cu the svmHuong dan cu the svm
Huong dan cu the svmtaikhoan262
 

Similar a Spring 2003 (20)

Problem 1 – First-Order Predicate Calculus (15 points)
Problem 1 – First-Order Predicate Calculus (15 points)Problem 1 – First-Order Predicate Calculus (15 points)
Problem 1 – First-Order Predicate Calculus (15 points)
 
Midterm
MidtermMidterm
Midterm
 
Homework Value of InformationPlease respond to the following.docx
Homework Value of InformationPlease respond to the following.docxHomework Value of InformationPlease respond to the following.docx
Homework Value of InformationPlease respond to the following.docx
 
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docx
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docxCSCI 2033 Elementary Computational Linear Algebra(Spring 20.docx
CSCI 2033 Elementary Computational Linear Algebra(Spring 20.docx
 
Digital electronics k map comparators and their function
Digital electronics k map comparators and their functionDigital electronics k map comparators and their function
Digital electronics k map comparators and their function
 
Mechanical Engineering Homework Help
Mechanical Engineering Homework HelpMechanical Engineering Homework Help
Mechanical Engineering Homework Help
 
3rd Semester Computer Science and Engineering (ACU-2022) Question papers
3rd Semester Computer Science and Engineering  (ACU-2022) Question papers3rd Semester Computer Science and Engineering  (ACU-2022) Question papers
3rd Semester Computer Science and Engineering (ACU-2022) Question papers
 
This quiz is open book and open notes/tutorialoutlet
This quiz is open book and open notes/tutorialoutletThis quiz is open book and open notes/tutorialoutlet
This quiz is open book and open notes/tutorialoutlet
 
Matlab 1
Matlab 1Matlab 1
Matlab 1
 
Quantum Computing Notes Ver 1.2
Quantum Computing Notes Ver 1.2Quantum Computing Notes Ver 1.2
Quantum Computing Notes Ver 1.2
 
IRJET- Solving Quadratic Equations using C++ Application Program
IRJET-  	  Solving Quadratic Equations using C++ Application ProgramIRJET-  	  Solving Quadratic Equations using C++ Application Program
IRJET- Solving Quadratic Equations using C++ Application Program
 
0013 chapter vi
0013 chapter vi0013 chapter vi
0013 chapter vi
 
3rd Semester Computer Science and Engineering (ACU) Question papers
3rd Semester Computer Science and Engineering  (ACU) Question papers3rd Semester Computer Science and Engineering  (ACU) Question papers
3rd Semester Computer Science and Engineering (ACU) Question papers
 
College Algebra MATH 107 Spring, 2016, V4.7 Page 1 of .docx
College Algebra   MATH 107   Spring, 2016, V4.7 Page 1 of .docxCollege Algebra   MATH 107   Spring, 2016, V4.7 Page 1 of .docx
College Algebra MATH 107 Spring, 2016, V4.7 Page 1 of .docx
 
Guide
GuideGuide
Guide
 
Huong dan cu the svm
Huong dan cu the svmHuong dan cu the svm
Huong dan cu the svm
 
A.B. .docx
A.B. .docxA.B. .docx
A.B. .docx
 
Guide
GuideGuide
Guide
 
MATH MODULE GRADE 8
MATH MODULE GRADE 8MATH MODULE GRADE 8
MATH MODULE GRADE 8
 
MATH MODULE GRADE 8
MATH MODULE GRADE 8MATH MODULE GRADE 8
MATH MODULE GRADE 8
 

Más de butest

EL MODELO DE NEGOCIO DE YOUTUBE
EL MODELO DE NEGOCIO DE YOUTUBEEL MODELO DE NEGOCIO DE YOUTUBE
EL MODELO DE NEGOCIO DE YOUTUBEbutest
 
1. MPEG I.B.P frame之不同
1. MPEG I.B.P frame之不同1. MPEG I.B.P frame之不同
1. MPEG I.B.P frame之不同butest
 
LESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALLESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALbutest
 
Timeline: The Life of Michael Jackson
Timeline: The Life of Michael JacksonTimeline: The Life of Michael Jackson
Timeline: The Life of Michael Jacksonbutest
 
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...butest
 
LESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALLESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALbutest
 
Com 380, Summer II
Com 380, Summer IICom 380, Summer II
Com 380, Summer IIbutest
 
The MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazz
The MYnstrel Free Press Volume 2: Economic Struggles, Meet JazzThe MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazz
The MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazzbutest
 
MICHAEL JACKSON.doc
MICHAEL JACKSON.docMICHAEL JACKSON.doc
MICHAEL JACKSON.docbutest
 
Social Networks: Twitter Facebook SL - Slide 1
Social Networks: Twitter Facebook SL - Slide 1Social Networks: Twitter Facebook SL - Slide 1
Social Networks: Twitter Facebook SL - Slide 1butest
 
Facebook
Facebook Facebook
Facebook butest
 
Executive Summary Hare Chevrolet is a General Motors dealership ...
Executive Summary Hare Chevrolet is a General Motors dealership ...Executive Summary Hare Chevrolet is a General Motors dealership ...
Executive Summary Hare Chevrolet is a General Motors dealership ...butest
 
Welcome to the Dougherty County Public Library's Facebook and ...
Welcome to the Dougherty County Public Library's Facebook and ...Welcome to the Dougherty County Public Library's Facebook and ...
Welcome to the Dougherty County Public Library's Facebook and ...butest
 
NEWS ANNOUNCEMENT
NEWS ANNOUNCEMENTNEWS ANNOUNCEMENT
NEWS ANNOUNCEMENTbutest
 
C-2100 Ultra Zoom.doc
C-2100 Ultra Zoom.docC-2100 Ultra Zoom.doc
C-2100 Ultra Zoom.docbutest
 
MAC Printing on ITS Printers.doc.doc
MAC Printing on ITS Printers.doc.docMAC Printing on ITS Printers.doc.doc
MAC Printing on ITS Printers.doc.docbutest
 
Mac OS X Guide.doc
Mac OS X Guide.docMac OS X Guide.doc
Mac OS X Guide.docbutest
 
WEB DESIGN!
WEB DESIGN!WEB DESIGN!
WEB DESIGN!butest
 

Más de butest (20)

EL MODELO DE NEGOCIO DE YOUTUBE
EL MODELO DE NEGOCIO DE YOUTUBEEL MODELO DE NEGOCIO DE YOUTUBE
EL MODELO DE NEGOCIO DE YOUTUBE
 
1. MPEG I.B.P frame之不同
1. MPEG I.B.P frame之不同1. MPEG I.B.P frame之不同
1. MPEG I.B.P frame之不同
 
LESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALLESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIAL
 
Timeline: The Life of Michael Jackson
Timeline: The Life of Michael JacksonTimeline: The Life of Michael Jackson
Timeline: The Life of Michael Jackson
 
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...
Popular Reading Last Updated April 1, 2010 Adams, Lorraine The ...
 
LESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIALLESSONS FROM THE MICHAEL JACKSON TRIAL
LESSONS FROM THE MICHAEL JACKSON TRIAL
 
Com 380, Summer II
Com 380, Summer IICom 380, Summer II
Com 380, Summer II
 
PPT
PPTPPT
PPT
 
The MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazz
The MYnstrel Free Press Volume 2: Economic Struggles, Meet JazzThe MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazz
The MYnstrel Free Press Volume 2: Economic Struggles, Meet Jazz
 
MICHAEL JACKSON.doc
MICHAEL JACKSON.docMICHAEL JACKSON.doc
MICHAEL JACKSON.doc
 
Social Networks: Twitter Facebook SL - Slide 1
Social Networks: Twitter Facebook SL - Slide 1Social Networks: Twitter Facebook SL - Slide 1
Social Networks: Twitter Facebook SL - Slide 1
 
Facebook
Facebook Facebook
Facebook
 
Executive Summary Hare Chevrolet is a General Motors dealership ...
Executive Summary Hare Chevrolet is a General Motors dealership ...Executive Summary Hare Chevrolet is a General Motors dealership ...
Executive Summary Hare Chevrolet is a General Motors dealership ...
 
Welcome to the Dougherty County Public Library's Facebook and ...
Welcome to the Dougherty County Public Library's Facebook and ...Welcome to the Dougherty County Public Library's Facebook and ...
Welcome to the Dougherty County Public Library's Facebook and ...
 
NEWS ANNOUNCEMENT
NEWS ANNOUNCEMENTNEWS ANNOUNCEMENT
NEWS ANNOUNCEMENT
 
C-2100 Ultra Zoom.doc
C-2100 Ultra Zoom.docC-2100 Ultra Zoom.doc
C-2100 Ultra Zoom.doc
 
MAC Printing on ITS Printers.doc.doc
MAC Printing on ITS Printers.doc.docMAC Printing on ITS Printers.doc.doc
MAC Printing on ITS Printers.doc.doc
 
Mac OS X Guide.doc
Mac OS X Guide.docMac OS X Guide.doc
Mac OS X Guide.doc
 
hier
hierhier
hier
 
WEB DESIGN!
WEB DESIGN!WEB DESIGN!
WEB DESIGN!
 

Spring 2003

  • 1. University of Wisconsin – Madison Computer Sciences Department CS 760 - Machine Learning Spring 2003 Exam 7:15-9:15pm, May 6, 2003 Room 3345 Engineering Hall CLOSED BOOK (one sheet of notes and a calculator allowed) Write your answers on these pages and show your work. If you feel that a question is not fully specified, state any assumptions that you need to make in order to solve the problem. You may use the backs of these sheets for scratch work. Write your name on this and all other pages of this exam. Make sure your exam contains 5 problems on 10 pages. Name ________________________________________________________________ Student ID ________________________________________________________________ Problem Score Max Score 1 ______ 40 2 ______ 17 3 ______ 10 4 ______ 12 5 ______ 21 TOTAL ______ 100
  • 2. Name: _______________________________________ Problem 1 – Learning from Labeled Examples (40 points) Imagine that you are given the following set of training examples. Feature F1 can take on the values a, b, or c; Feature F2 is Boolean-valued; and Feature F3 is always a real-valued number in [0,1]. F1 F2 F3 Category Example 1 a T 0.2 + Example 2 b F 0.5 + Example 3 b F 0.9 + Example 4 b T 0.6 – Example 5 a F 0.1 – Example 6 a T 0.7 – a) How might a Naive Bayes system classify the following test example? Be sure to show your work. (Discretize the numeric feature into three equal-width bins.) F1 = c F2 = T F3 = 0.8 b) Describe how a 2-nearest-neighbor algorithm might classify Part a’s test example. Page 2 of 10
  • 3. Name: _______________________________________ c) Show the calculations that ID3 would perform to determine the root node of a decision tree using the above training examples. d) Using the “decision stump” learner as boosting’s weak learner, what would the weights be on each of the six training examples after one (1) round of AdaBoost? Show your work. Weight on Example 1: _________ Weight on Example 2: _________ Weight on Example 3: _________ Weight on Example 4: _________ Weight on Example 5: _________ Weight on Example 6: _________ e) Show the feature space for this training set. Use only features F2 and F3 (ie, for this part, ignore F1). Show what a linear support-vector machine might learn from this training set. Briefly explain your answer. Page 3 of 10
  • 4. Name: _______________________________________ f) Create a neural network with no hidden units and a sigmoidal output unit. Initialize all the free parameters to 2 and use a learning rate of 0.1. Use squared error is the error function. i. Draw a picture of this neural network before training. ii. Calculate ∂ Error / ∂ weighti for this network. Recall that the derivative of the sigmoid is output × (1 – output). iii. Show how the free parameters would change after backpropagating on the first training example in the table above. Indicate the weight changes on the neural network you drew above and show your work below. iv. According to Rumelhart, which error function should one use on classification tasks such as this one? And how should one interpret the neural network’s output? Page 4 of 10
  • 5. Name: _______________________________________ Problem 2 – Reinforcement Learning (17 points) Consider the deterministic reinforcement environment drawn below. The numbers on the arcs indicate the immediate rewards. Let the discount rate equal 0.5. r = -10 -15 sta a b rt 15 6 10 -6 -3 3 -2 c 5 en d 0 Assume that the current Q table is represented by the Q values on the arcs on the environment's state-action graph below. Also assume that the learner is currently in Node b. Q = -1 -7 sta a b rt 14 7 12 -1 -9 1 -3 c 5 en d 0 a) The learner chooses to do an exploration step out of Node b, using the “softmax” function. What is the probability is that the learner will choose the action that leads to Node a (be sure to state any assumptions you need to make in order to compute this probability)? Page 5 of 10
  • 6. Name: _______________________________________ b) Assume that the learner is magically transported to Node c and that the Q table is still in the state shown below. For this question, assume the learner chooses to first go to Node b and then to Node a. Be sure to show the calculations that produce your answers to the questions below. Q = -1 -7 sta a b rt 14 7 12 -1 -9 1 -3 c 5 en d 0 i. Using one-step Q learning, what new value would there be on the c→ b arc? ii. Using one-step SARSA learning, what new value would there be on the c→ b arc? c) This time assume the learner is currently in Node start and the Q table contains the values in the graph above (the values before you answered Part b). The learner does three (3) exploitation steps and employs two-step Q learning along the way. Indicate on the graph above what changes in the Q table this time. Show your calculations below. Page 6 of 10
  • 7. Name: _______________________________________ Problem 3 – Computational Learning Theory (10 points) a) Assume that we have the predictions below of five experts, as well as the correct answer. Using the weighted majority algorithm (with β=0.5) in order to track the best expert, show how the weight given to each expert changes after each example. Show your work. Expert 1 2 3 4 5 Correct Answer T T T F F F Weights ________________________________________ F T F T T T Weights ________________________________________ T F F F T F Weights ________________________________________ b) Consider the Boolean concept class of circles in the space of two numeric features (points on or inside the circle are considered positive). What is the VC-dimension for this concept space? Explain. Page 7 of 10
  • 8. Name: _______________________________________ Problem 4 – Overfitting Avoidance (12 points) a) Use a Venn diagram of feature space to illustrate how a decision tree might overfit a training set. Briefly explain your answer. b) Briefly describe how one can choose the number of hidden units to use in a neural network in order to reduce the chances of overfitting. c) Informally (i.e., verbally or pictorially) explain how using a kernel might lead to overfitting in a support-vector machine. d) Outline a methodology where an experimenter incorrectly uses cross validation to set parameters and estimate future accuracy. Page 8 of 10
  • 9. Name: _______________________________________ Problem 5 – Short Discussion Questions (21 points) a) Using a Venn diagram, qualitatively illustrate how 1-NN (one nearest-neighbor) partitions feature space. Assume a Boolean classification task. Briefly explain your answer. b) Consider creating an ensemble of eleven support-vector machines (SVM’s) using bagging and a standard linear SVM. Dataset A has 100 support vectors and Dataset B has 1000 support vectors (when run on all the data); both datasets contain 10,000 examples and on both datasets the 10-fold cross-validation accuracy of a single SVM is 80%. On which dataset would you expect bagging to improve accuracy the most? Why? c) What does transduction mean in machine learning? Page 9 of 10
  • 10. Name: _______________________________________ d) Explain the role of slack variables in support vector machines. e) What do you feel is the most important improvement that PAC learning offers over learning in the limit? Explain your answer. f) Noah Taul does a proper 10-fold cross-validation of his new algorithm compared to backprop. A paired t-test shows he cannot disregard the null hypothesis. What is the null hypothesis? And how does a t-test allow one to address this hypothesis? Have a good vacation! Page 10 of 10