Information for CS 536: Machine Learning, Fall 2002

Lectures take place 4:30-5:50pm in CoRE 301.

Instructor:

Professor Haym Hirsh
CoRE 317
732-445-4176
h i r s h (a t) c s . r u t g e r s . e d u
Office hours: Tuesdays/Thursday 1:00-2:00pm

Teaching Assistant:

Eiman Elnahrawy
CoRE 246
732-445-4714
e i m a n (a t) p a u l . r u t g e r s . e d u
Office hours: Fridays 10:30am-12:30pm

• Lecture 1, Monday, September 9: Introduction
  Course Information (latest version: Sep 17 8:00pm)
• Lecture 2, Wednesday, September 11: Concept Learning
  Problem Set 1, Due Sep 18 (latest version: Sep 11 2:00pm)
• Lecture 3, Wednesday, September 18: Cross-Validation
  Problem Set 1 Due
• Lecture 4, Monday, September 23: Learning Decision Trees
  Problem Set 2, Due Sep 30 (latest version: Sep 23 4:00pm)
• Lecture 5, Wednesday, September 25: Learning Decision Trees
• Lecture 6, Monday, September 30: Learning Decision Trees; Naive Bayes
  Problem Set 2 Due
• Lecture 7, Wednesday, October 2: Naive Bayes
  Reading Assignment: William Cohen, "Fast Effective Rule Induction", in Proceedings of the Twelfth International Conference on Machine Learning (1995)
• Lecture 8, Monday, October 7: Evaluation; Neural Networks
• Lecture 9, Wednesday, October 9: Neural Networks
  Problem Set 3, Due Oct 16 (latest version: Oct 9 1:00pm)
  Programming Assignment 1, Due Oct 23 (latest version: Oct 16 10:00pm)
  Do not use the following data sets:
  • balance-scale
  • breast-w
  • colic.ORIG
  • colic
  • diabetes
  • glass
  • heart-statlog
  • ionosphere
  • iris
  • labor
  • letter
  • segment
  • sonar
  • soybean
  • vehicle
  • vote
  • vowel
  • waveform
  For the other data sets use Weka to delete the numeric attributes and all attributes with missing values: Under "Preprocess", deselect the numeric attributes, and then click on "Apply Filters".
• Lecture 10, Monday, October 14: Evaluation
• Lecture 11, Wednesday, October 16: Evaluation
• Lecture 12, Monday, October 21: Evaluation; Rule Learning
• Lecture 13, Wednesday, October 23: Rule Learning
• Midterm, Monday, October 28
  Sample Midterm
  Sample Midterm Solutions
  Topics:
  • 1-nearest neighbor
  • 1R
  • Decision trees
    • Divide and conquer algorithm
    • Gain ratio attribute selection criterion
    • Handling numeric splits
    • Handling missing values
    • Pruning methods (reduced error, pessimistic error)
  • Naive Bayes
    • Laplace smoothing
    • Handling numerical values
    • Handling missing values
  • Neural networks
    • Perceptrons
    • Linear separability
    • Multi-layer networks
    • Sigmoidal activation units
    • Backpropagation training rule
  • Rule learning: RIPPER
  • Evaluation
    • Cross-validation
    • Stratified sampling
    • Confidence intervals
    • t-tests
    • Sign test
    • Scatter plots
    • Quatratic and informational loss
    • Confusion matrices
    • FP, TP, FN, TN, precision, recall
    • ROC curves
    • Lift charts
    • Precision-recall curves
• Lecture 14, Wednesday, October 30: Setting Parameters; Instance-Based Learning
  More about setting parameteres with cross-validation can be found in the 1993 paper by Cullen Schaffer, "Selecting a Classification Method by Cross-Validation"
• Lecture 15, Monday, November 4: Instance-Based Learning
• Lecture 16, Wednesday, November 6: Learning Association Rules
  Guest lecture by Professor Tomasz Imielinski
• Lecture 17, Monday, November 11: Version Spaces
  Problem Set 4, Due Nov 18 (latest version: Nov 11 4:30pm)
• Lecture 18, Wednesday, November 13: Version Spaces
• Lecture 19, Monday, November 18: Learning Bayes Nets
  Guest lecture by David Madigan
  Problem Set 5, Due Nov 25 (latest version: Nov 18 6:15pm)
• Lecture 20, Wednesday, November 20: Neural Nets for Branch Prediction
  Guest lecture by Daniel Jimenez
  Lecture notes
• Lecture 21, Monday, November 25: Version Spaces
  Problem Set 6, Due Dec 2 (latest version: Nov 25 4:30pm)
  Data for Problem Set 6
• No Lecture, Wednesday, November 27: Thanksgiving
• Lecture 22, Monday, December 2: Reinforcement Learning
  Guest lecture by Michael Littman
  Lecture notes
• Lecture 23, Wednesday, December 4: Reinforcement Learning
  Guest lecture by Michael Littman
  Lecture notes
  Problem Set 7, Due Dec 9 to get graded by the last lecture, due Dec 11 otherwise (latest version: Nov 25 6:30pm)
  Data for Problem Set 7
• Lecture 24, Monday, December 9: PAC Learning
• Lecture 25, Wednesday, December 11: Clustering
  Problem Set 8, due not submit, this is to help you prepare for the final exam (latest version: Dec 11 12:30pm)
• Final Exam, Monday, December 16, 4-7pm

Final Grades are now available, listed by last four digits of your ID number.

Course Textbooks:

• Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations by Ian H. Witten and Eibe Frank, published in 2000 by Morgan Kaufmann Publishers.
• Machine Learning by Tom Mitchell, published in 1997 by McGraw Hill.

There will be some interesting classes offered next semester, including:

• Computer Vision, Vladimir Pavlovic
• Computers in Biomedicine, Casimir Kulikowski
• Conceptual Modeling and Ontologies in Computer Science, Alex Borgida
• Massive Data Set Processing, S. Muthukrishnan
• Multisensory Modeling, Simulation, and Interaction, Dinesh Pai
• Natural Language, Matthew Stone
• Seminar in Computer Vision: Looking at People, Ahmed Elgammal
• Seminar in Computer Architecture, Daniel Jiminez
• Seminar on Medical Applications and Computer Vision, Dimitris Metaxas