CS425/528: Schedule/Readings/Notes

CS425/528: Machine Learning

Schedule/Readings/Notes
Fall 2010

Subject to change. Check back frequently for updates.
Last updated: November 23, 2010

Date Topics Assigned Readings

Thurs. 8/19 Course Introduction
Intro. to Learning
Lecture slides

Tues. 8/24 Designing learning systems

(No lecture slides, but see Mitchell's Ch. 1 slides for related material, especially slides 11-21.)
Ch. 1 of Mitchell or Ch. 1 of Nilsson

Thurs. 8/26 Neural Networks
Introduction
Perceptrons
Gradient descent

(No lecture slides, but see Mitchell's Ch. 4 slides for related material.)
Ch. 4 of Mitchell or Ch. 4 of Nilsson

"The Basic Ideas in Neural Networks", by Rumelhart et al., Communications of the ACM, 37(3): 87-92, 1994.

Tues. 8/31 Neural Networks (con't.)
Stochastic gradient descent
Multi-layer feedforward neural networks
Ch. 4 of Mitchell or Ch. 4 of Nilsson

Handout: Notes on Neural Networks

Thurs. 9/2 Neural Networks (con't.)
Back Propagation
Convergence issues
K-fold cross-validation
Ch. 4. of Mitchell or Ch. 4 of Nilsson

Tues. 9/07 Neural Networks (con't.)
Momentum
Design and performance issues
Discussion of project #1
Ch. 4 of Mitchell or Ch. 4 of Nilsson
Project 1 assigned

Thurs. 9/09 Neural Networks (con't.)
Case studies
Evaluation of learning algorithms:
Sensitivity, specificity
Type I and Type II errors
Confusion matrix
ROC curve

Tutorial on LaTex

Case Studies:
A case study of using artificial neural networks for classifying cause of death from verbal autopsy, International Journal of Epidemiology, 2001, 30:515-520.
Predicting Students' Academic Performance using Artificial Neural Networks: A Case Study of an Engineering Course, The Pacific Journal of Science and Technology, Vol. 9, No. 1, May-June 2008, pgs. 72-79.

Analysis of Hidden Units in a Layered Network Trained to Classify Sonar Targets, by Gorman and Sejnowski, Neural Networks, Vol. 1, pgs. 75-89, 1988.
A case study on using neural networks to perform technical forecasting of forex, Neurocomputing, Vol. 24, pgs. 79-98, 2000.
Mangerial applications of neural networks: The Case of Bank Failure Predictions, Management Sceince, Vol. 38, No. 7, July 1992, pgs. 926-947.

Tues. 9/14 Bias in machine learning

Genetic Algorithms
   Introduction
   Representation
   Crossover operators

Evolutionary programming example: GOLEM project for evolving robots
(No lecture slides, but see Mitchell's Ch. 9 slides for related material.)
Ch. 9.1-9.2 of Mitchell

Thurs. 9/16 Genetic Algorithms
   Selection/Fitness
   TSP example

Ch. 9.2-9.3 of Mitchell

Tues. 9/21 Genetic Programming
   Introduction
   Example: Boolean 11-Multiplexer Ch. 9.5 of Mitchell
Extra handouts: excerpts from John Koza's Genetic Programming, parts of Ch. 6 (pgs. 79-94) and Ch. 7 (pgs. 147-187)

Thurs. 9/23 Genetic Programming
   Closure and Sufficiency
   Generating initial population
   Example: Santa Fe Ant
Koza handouts from last class

Tues. 9/28 Genetic Programming
   Example: Symbolic regression
   Example: Blocks world
   Dealing with "code bloat"

Revisiting GOLEM example

Automatic design and manufacture of robotic lifeforms, Nature, vol. 406, August 31, 2000, pg. 974-978.

Thurs. 9/30 Discussion of Project #2
Introduction to Reinforcement Learning
Project #2 Assigned
Ch. 13.1 - 13.2 of Mitchell
Ch. 1 of Sutton and Barto text on RL

Tues. 10/05 In-class design exercise
(no lecture; instructor on travel)

Thurs. 10/07 Fall break, no class

Tues. 10/12 Reinforcement learning
   Markov Property
   Markov Decision Processes
   K-Armed Bandit example
   Q-Learning Ch. 13.3 of Mitchell
Ch. 3, 6.5 of Sutton and Barto text on RL

Thurs. 10/14 Reinforcement learning
   TD Learning
   Sarsa (lambda)
   Eligibility traces Ch. 13.4 - 13.5 of Mitchell
Ch. 6 and 7 of Sutton and Barto text on RL

Tues. 10/19 (Guest lecture by GTA; instructor on research travel)
Tutorial on Player/Stage
Reinforcement learning (con't.)
   Epsilon greedy action selection
   RL in-class exercise Paper introducing Player
Player/Stage Getting Started Guide (UTK-specific)
Player/Stage Documentation (Public-domain website)
Mini-Assignment #1 Assigned (easy!!)

Thurs. 10/21 (Guest lecture by GTA; instructor on research travel)
Reinforcement learning (con't.)
   Case Studies:
     TD-Gammon
     Checkers
     Elevator Dispatch
Ch. 11 of Sutton and Barto text on RL

Tues. 10/26 Discussion of Project #3
Wrapup of Reinforcement Learning
Introduction to Support Vector Machines (SVMs)

Handout: pages 744-748 of Artificial Intelligence, 3rd edition, by Russell and Norvig, Prentice-Hall, 2010.
Project #3 assigned

Thurs. 10/28 SVMs (con't.)
   Intro to convex optimization
   Intro to quadradic programming
   Intro to concepts of primal/dual
   SVM dual representation

Tues. 11/02 SVMs (con't.)
   Maximizing margin formulation Handout: pages 73-87 of Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.

Thurs. 11/04 SVMs (con't.)
   Nonlinear SVMs
   Kernels and the Kernel Trick
   Soft margin classifiers Handout: pages 102-109, 114-122 of Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.

Tues. 11/09 No lecture; class optional
Q&A on Project #3, led by GTA

Thurs. 11/11 SVMs (con't.)
   Multi-Class Classification
   Regression with SVMs Handout: pages 185-202 Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.

Tues. 11/16 SVMs (con't.)
   Practical issues:
      Categorical features
      Data scaling
      Missing data
      Model selection (use RBFs)
   Parameter selection using cross-validation and grid search
AdaBoosting and connection to SVMs
AdaBoosting with SVMs for unbalanced data Handout: A practical guide to support vector classification, by Hsu, et al.
Handout: A Short Introduction to Boosting, by Freund and Schapire
Handout: AdaBoost with SVM-based component classifiers, by Li, Wang, and Sung

Thurs. 11/18 AdaBoosting with SVMs for unbalanced data (con't.)
Introduction to LIBSVM
Discussion of Project #4 LIBSVM website

Handout: AdaBoost.M1 description, by Freund and Schapire

Project #4 assigned

Tues. 11/23 Stable/unstable learners
Inductive bias
Combining learners -- bagging Background reading: A simple algorithm for learning stable machines, by Andonova, et al.

Thurs. 11/25 Happy Thanksgiving! (no class)

Tues. 11/30 (Last class; no final exam)
No Free Lunch theorem
Combining learners (con't.)
Course wrapup

Date	Topics	Assigned Readings
Thurs. 8/19	Course Introduction Intro. to Learning Lecture slides
Tues. 8/24	Designing learning systems (No lecture slides, but see Mitchell's Ch. 1 slides for related material, especially slides 11-21.)	Ch. 1 of Mitchell or Ch. 1 of Nilsson
Thurs. 8/26	Neural Networks Introduction Perceptrons Gradient descent (No lecture slides, but see Mitchell's Ch. 4 slides for related material.)	Ch. 4 of Mitchell or Ch. 4 of Nilsson "The Basic Ideas in Neural Networks", by Rumelhart et al., Communications of the ACM, 37(3): 87-92, 1994.
Tues. 8/31	Neural Networks (con't.) Stochastic gradient descent Multi-layer feedforward neural networks	Ch. 4 of Mitchell or Ch. 4 of Nilsson Handout: Notes on Neural Networks
Thurs. 9/2	Neural Networks (con't.) Back Propagation Convergence issues K-fold cross-validation	Ch. 4. of Mitchell or Ch. 4 of Nilsson
Tues. 9/07	Neural Networks (con't.) Momentum Design and performance issues Discussion of project #1	Ch. 4 of Mitchell or Ch. 4 of Nilsson Project 1 assigned
Thurs. 9/09	Neural Networks (con't.) Case studies Evaluation of learning algorithms: Sensitivity, specificity Type I and Type II errors Confusion matrix ROC curve Tutorial on LaTex	Case Studies: A case study of using artificial neural networks for classifying cause of death from verbal autopsy, International Journal of Epidemiology, 2001, 30:515-520. Predicting Students' Academic Performance using Artificial Neural Networks: A Case Study of an Engineering Course, The Pacific Journal of Science and Technology, Vol. 9, No. 1, May-June 2008, pgs. 72-79. Analysis of Hidden Units in a Layered Network Trained to Classify Sonar Targets, by Gorman and Sejnowski, Neural Networks, Vol. 1, pgs. 75-89, 1988. A case study on using neural networks to perform technical forecasting of forex, Neurocomputing, Vol. 24, pgs. 79-98, 2000. Mangerial applications of neural networks: The Case of Bank Failure Predictions, Management Sceince, Vol. 38, No. 7, July 1992, pgs. 926-947.
Tues. 9/14	Bias in machine learning Genetic Algorithms Introduction Representation Crossover operators Evolutionary programming example: GOLEM project for evolving robots (No lecture slides, but see Mitchell's Ch. 9 slides for related material.)	Ch. 9.1-9.2 of Mitchell
Thurs. 9/16	Genetic Algorithms Selection/Fitness TSP example	Ch. 9.2-9.3 of Mitchell
Tues. 9/21	Genetic Programming Introduction Example: Boolean 11-Multiplexer	Ch. 9.5 of Mitchell Extra handouts: excerpts from John Koza's Genetic Programming, parts of Ch. 6 (pgs. 79-94) and Ch. 7 (pgs. 147-187)
Thurs. 9/23	Genetic Programming Closure and Sufficiency Generating initial population Example: Santa Fe Ant	Koza handouts from last class
Tues. 9/28	Genetic Programming Example: Symbolic regression Example: Blocks world Dealing with "code bloat" Revisiting GOLEM example	Automatic design and manufacture of robotic lifeforms, Nature, vol. 406, August 31, 2000, pg. 974-978.
Thurs. 9/30	Discussion of Project #2 Introduction to Reinforcement Learning	Project #2 Assigned Ch. 13.1 - 13.2 of Mitchell Ch. 1 of Sutton and Barto text on RL
Tues. 10/05	In-class design exercise (no lecture; instructor on travel)
Thurs. 10/07	Fall break, no class
Tues. 10/12	Reinforcement learning Markov Property Markov Decision Processes K-Armed Bandit example Q-Learning	Ch. 13.3 of Mitchell Ch. 3, 6.5 of Sutton and Barto text on RL
Thurs. 10/14	Reinforcement learning TD Learning Sarsa (lambda) Eligibility traces	Ch. 13.4 - 13.5 of Mitchell Ch. 6 and 7 of Sutton and Barto text on RL
Tues. 10/19	(Guest lecture by GTA; instructor on research travel) Tutorial on Player/Stage Reinforcement learning (con't.) Epsilon greedy action selection RL in-class exercise	Paper introducing Player Player/Stage Getting Started Guide (UTK-specific) Player/Stage Documentation (Public-domain website) Mini-Assignment #1 Assigned (easy!!)
Thurs. 10/21	(Guest lecture by GTA; instructor on research travel) Reinforcement learning (con't.) Case Studies: TD-Gammon Checkers Elevator Dispatch	Ch. 11 of Sutton and Barto text on RL
Tues. 10/26	Discussion of Project #3 Wrapup of Reinforcement Learning Introduction to Support Vector Machines (SVMs)	Handout: pages 744-748 of Artificial Intelligence, 3rd edition, by Russell and Norvig, Prentice-Hall, 2010. Project #3 assigned
Thurs. 10/28	SVMs (con't.) Intro to convex optimization Intro to quadradic programming Intro to concepts of primal/dual SVM dual representation
Tues. 11/02	SVMs (con't.) Maximizing margin formulation	Handout: pages 73-87 of Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.
Thurs. 11/04	SVMs (con't.) Nonlinear SVMs Kernels and the Kernel Trick Soft margin classifiers	Handout: pages 102-109, 114-122 of Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.
Tues. 11/09	No lecture; class optional Q&A on Project #3, led by GTA
Thurs. 11/11	SVMs (con't.) Multi-Class Classification Regression with SVMs	Handout: pages 185-202 Knowledge Discovery with Support Vector Machines, by Lutz Hamel, John Wiley and Sons, Inc., 2009.
Tues. 11/16	SVMs (con't.) Practical issues: Categorical features Data scaling Missing data Model selection (use RBFs) Parameter selection using cross-validation and grid search AdaBoosting and connection to SVMs AdaBoosting with SVMs for unbalanced data	Handout: A practical guide to support vector classification, by Hsu, et al. Handout: A Short Introduction to Boosting, by Freund and Schapire Handout: AdaBoost with SVM-based component classifiers, by Li, Wang, and Sung
Thurs. 11/18	AdaBoosting with SVMs for unbalanced data (con't.) Introduction to LIBSVM Discussion of Project #4	LIBSVM website Handout: AdaBoost.M1 description, by Freund and Schapire Project #4 assigned
Tues. 11/23	Stable/unstable learners Inductive bias Combining learners -- bagging	Background reading: A simple algorithm for learning stable machines, by Andonova, et al.
Thurs. 11/25	Happy Thanksgiving! (no class)
Tues. 11/30	(Last class; no final exam) No Free Lunch theorem Combining learners (con't.) Course wrapup

CS425/528: Machine Learning

Schedule/Readings/Notes Fall 2010

Schedule/Readings/Notes
Fall 2010