CS 4780 - Machine Learning
Machine Learning

CS 4780/5780 - Fall 2012
Cornell University
Department of Computer Science

 
Time and Place
First lecture: August 23, 2012
Last lecture: November 29, 2012
  • Tuesday, 1:25pm - 2:40pm in Olin 255
  • Thursday, 1:25pm - 2:40pm in Olin 255

First Prelim Exam: October 16
Second Prelim Exam: November 20
Project Due Date: December 13

Instructor
Thorsten Joachims (homepage)
Online Resources
  • [Piazza Forum]  We greatly prefer that you use this Piazza forum for questions and discussions.  The forum is monitored all the TA's and the prof -- you will get the best response time. And all the TA's will know the question you asked and the answers you receive.
  • [CMS] Assignments, the prereq exam, and all grades are posted on CMS.
  • [Videonote] Recordings of last year's lectures with annotations. Lectures this year will be somewhat different, but this can still be a useful reference.
Teaching Assistants and Consultants
Igor Labutov, TA (homepage)
Moontae Lee, TA (homepage)
Joshua Moore, TA (homepage)
Harry Terkelsen, Consultant
Declan Boyd, Consultant
Jason Zhao, Consultant
Joe Mongeluzzi, Consultant
Kyle Hsu, Consultant
Joseph Staehle, Consultant
Emma Kilfoyle, Consultant
Office Hours
Time Person Location
Monday, 3:30 to 4:30 Moontae Lee Upson 328 Bay A
Monday, 4:30 to 5:30 Harry Terkelsen Upson 328 Bay A
Tuesday, 4:00 to 5:00 Declan Boyd Upson 328 Bay A
Tuesday, 5:00 to 6:00 Josh Moore Upson 5162
Tuesday, 6:00 to 7:00 Igor Labutov Upson 328 Bay B
Wednesday, 3:30 to 4:30 Moontae Lee Upson 328 Bay A
Wednesday, 5:00 to 6:00 Joey Staehle Upson 328 Bay A
Wednesday, 6:00 to 7:00 Igor Labutov Upson 328 Bay B
  Thursday, 2:40 to 4:00 Thorsten Joachims Upson 4153
  Friday, 3:00 to 4:00 Josh Moore Upson 5162
  Friday, 4:00 top 5:00 Harry Terkelsen Upson 328 Bay A
  Saturday, 4:00 to 5:00 Emma Kilfoyle Upson 328 Bay A
Sunday, 4:00 to 5:00 Declan Boyd Upson 328 Bay A
Syllabus
Machine learning is concerned with the question of how to make computers learn from experience. The ability to learn is not only central to most aspects of intelligent behavior, but machine learning techniques have become key components of many software systems. For examples, machine learning techniques are used to create spam filters, to analyze customer purchase data, to understand natural language, or to detect fraudulent credit card transactions.

This course will introduce the fundamental set of techniques and algorithms that constitute machine learning as of today, ranging from classification methods like decision trees and support vector machines, over structured models like hidden Markov models, to clustering and matrix factorization methods for recommendation. The course will not only discuss individual algorithms and methods, but also tie principles and approaches together from a theoretical perspective. In particular, the course will cover the following topics: 

  • Concept Learning : Hypothesis space, version space
  • Instance-based Learning : K-Nearest Neighbors, collaborative filtering
  • Decision Trees : TDIDT, attribute selection, pruning and overfitting
  • ML Experimentation : Hypothesis tests, resampling estimates
  • Linear Rules : Perceptron, duality, mistake bound
  • Support Vector Machines : Optimal hyperplane, kernels, stability
  • Generative Models : Na´ve Bayes, linear discriminant analysis
  • Hidden Markov Models: probabilistic model, estimation, Viterbi
  • Structured Output Prediction : predicting sequences, rankings, etc.
  • Learning Theory : PAC learning, mistake bounds, VC dimension
  • Clustering : HAC, k-means, mixture of Gaussians
  • Recommendation Systems: Similarity based methods, matrix factorization

 

Slides and Handouts
08/23: Introduction (PDF)
  08/28: Instance-Based Learning (PDF)
  08/30: Decision Tree Learning (PDF)
  09/06: Prediction and Overfitting (PDF)
  09/11: Model Selection and Assessment (PDF)
  09/13: Linear Classifiers and Perceptrons (PDF)
  09/20: Optimal Hyperplanes and Support Vector Machines (PDF)
  09/25: SVM Duality and Leave-one-out Bounds (PDF)
  10/02: Kernels (PDF)
  10/04: Learning to Rank (PDF)
  10/11: Recommender Systems and Matrix Decomposition (PDF)
  10/23: Learning through Generative Modeling (PDF)
  10/30: Sequence Prediction and Hidden Markov Models (PDF)
  11/06: Statistical Learning Theory (PDF)
  11/13: Clustering (PDF)

Reference Material
The main textbooks for the class are
  • Tom Mitchell, "Machine Learning", McGraw Hill, 1997.
  • CS4780 course packet available at the Cornell Bookstore.

An additional textbook that can serve as a brief secondary reference on many topics in this class is

Ethem Alpaydin, "Introduction to Machine Learning", MIT Press, 2004.

The reading in the course packet are taken from the following books. In addition, these are some books for further reading beyond the scope of the course:

  • Cristianini, Shawe-Taylor, "Introduction to Support Vector Machines", Cambridge University Press, 2000. (online via Cornell Library) Schoelkopf, Smola, "Learning with Kernels", MIT Press, 2001. (online)
  • Bishop, "Pattern Recognition and Machine Learning", Springer, 2006.
  • Devroye, Gyoerfi, Lugosi, "A Probabilistic Theory of Pattern Recognition", Springer, 1997.
  • Duda, Hart, Stork, "Pattern Classification", Wiley, 2000.
  • Hastie, Tibshirani, Friedman, "The Elements of Statistical Learning", Springer, 2001.
  • Joachims, "Learning to Classify Text using Support Vector Machines", Kluwer, 2002.
  • Leeds Tutorial on HMMs (online)
  • Manning, Schuetze, "Foundations of Statistical Natural Language Processing", MIT Press, 1999. (online via Cornell Library)
  • Manning, Raghavan, Schuetze, "Introduction to Information Retrieval", Cambridge, 2008. (online)
  • Vapnik, "Statistical Learning Theory", Wiley, 1998.
Prerequisites
Programming skills (e.g. CS 2110 or CS 3110), and basic knowledge of linear algebra (e.g. MATH 2940) and probability theory (e.g. CS 2800).
Grading
This is a 4-credit course. Grades will be determined based on two written exams, a final project, homework assignments, and class participation.
  • 40%: 2 Prelim Exams
  • 15%: Final Project
  • 35%: Homework (~5 assignments)
  • 5%: Quizzes (in class)
  • 2%: Prereq Exam
  • 3%: Class Participation

To eliminate outlier grades for homeworks and quizzes, the lowest grade is replaced by the second lowest grade when grades are cumulated at the end of the semester.

All assignments are due at the beginning of class on the due date. Assignments turned in late will be charged a 1 percentage point reduction of the cumulated final homework grade for each period of 24 hours for which the assignment is late. However, every student has a budget of 5 late days (i.e. 24 hour periods after the time the assignment was due) throughout the semester for which there is no late penalty. So, if you have perfect scores of 100 on all 5 homeworks and a total of 7 late days, you final homework score will be 98 (which then accounts for 35% of your course grade). No assignment will be accepted after the solution was made public, which is typically 4 days after the time it was due. You can submit late assignments in class, in office hours, or to the office of a TA.

Graded homework assignments and prelims can be picked up in Upson 305 (note the change of room, opening hours Monday - Thursday 12:00pm - 4:00pm, Friday: 12:30pm - 4:00pm). Regrade requests can be submitted within 7 days after the grades have been made available on CMS. Regrade requests have to be submitted in writing and in hardcopy using this form (or similar). They can be submitted in class, in office hours, or to the office of a TA.

We always appreciate interesting homework solutions that go beyond the minimum. To reward homework solutions that are particularly nice, we will give you "Bonus Points". Bonus points are collected in a special category on CMS. Bonus points are not real points and are not summed up for the final grade, but they can nudge somebody to a higher grade who is right on the boundary.

All assignment, exam, and final grades are roughly on the following scale: A=92-100; B=82-88; C=72-78; D=60-68; F= below 60

Academic Integrity
This course follows the Cornell University Code of Academic Integrity. Each student in this course is expected to abide by the Cornell University Code of Academic Integrity. Any work submitted by a student in this course for academic credit will be the student's own work. Violations of the rules (e.g. cheating, copying, non-approved collaborations) will not be tolerated.

We run automatic cheating detection to detect violations of the collaboration rules.