CS281: Advanced Machine Learning

Announcements

• 2 December 2013: The final project poster session will be on the ground and first floors of Maxwell-Dworkin on Thursday 5 December from 2-4pm. Come a bit early to set things up.
• 3 November 2013: Assignment 5 is now available.
• 25 October 2013: Assignment 4 is now available.
• 18 October 2013: A practice midterm is available.
• 11 October 2013: A list of midterm study topics is now available.
• 6 October 2013: Assignment 3 is now available. The due date is extended to October 20.
• 21 September 2013: Assignment 2 is now available.
• 13 September 2013: Section times and places, as well as office hours, are now available.
• 10 September 2013: Please identify your section preferences here.
• 9 September 2013: Due to the size of the class, we are moving to a larger room: G115.
• 8 September 2013: If you filled out the survey, you should have now received notification of your status. If you have been assigned a place but do not intend to take the course, please let the staff know ASAP so that another student can take the course.
• 6 September 2013: Assignment 1 is now available.
• 5 September 2013: It has become apparent that cross-registered students will not be able to access the online quizzes until after study card day. As such, there will not be a quiz for the second lecture.
• 3 September 2013: Please fill out the survey.
• 31 August 2013: The syllabus is now available.
• 28 August 2013: Sign up on the Piazza discussion site.

• Assignment 1: Out 6 September 2013; Due 20 September 2013 [ assignment-1.pdf | assignment-1.tex ]
  

  Jester Data: These data are approximately 1.7 million ratings in the range [-10,10] of 150 jokes from 63,974 users. These data are from the Eigentaste Project at Berkeley. I have munged the data somewhat, so use the local copies here:

  • jester_ratings.dat.gz: Each row is formatted as [User ID] [Joke ID] [Rating]
  • jester_items.dat.gz: Maps the joke IDs to the joke text.
  • jester_items.clean.dat.gz: Maps the joke IDs to the joke text. (No punctuation, lowercased.)
  • jester_ratings_users.mat: Ratings chopped up into a cell array with one entry per user.
  • jester_ratings_jokes.mat: Ratings chopped up into a cell array with one entry per joke.
  The correct reference for these data is Ken Goldberg, Theresa Roeder, Dhruv Gupta and Chris Perkins. Eigentaste: A Constant Time Collaborative Filtering Algorithm. Information Retrieval 4(2):133-151, 2001. [pdf]

• Assignment 2: Out 21 September 2013; Due 4 October 2013 [ assignment-2.pdf | assignment-2.tex ]
  

  Spam Email Data: These data are 3000 training and 1601 test emails. Each has 57 features and a binary label in the last column. You can read more about the data here.

  • spam.train.dat.gz
  • spam.test.dat.gz

• Assignment 3: Out 4 October 2013; Due 18 October 2013 20 October 2013 [ assignment-3.pdf | assignment-3.tex ]
  

  Uses Jester data above.

• Assignment 4: Out 25 October 2013; Due 8 November 2013 [ assignment-4.pdf | assignment-4.tex ]
• Assignment 5: Out 8 November 2013; Due 22 November 2013 [assignment-5.pdf | assignment-5.tex ]

• Proposal: Due 25 October 2013
• Abstract and Status Report: Due 22 November 2013
• Poster Session: 5 December 2013 (2-4pm in Maxwell-Dworking)
• Final Report: Due 11 December 2013

• Required: Kevin Murphy, Machine Learning: A Probabilistic Perspective, MIT Press.
• Recommended: Christopher M. Bishop, Pattern Recognition and Machine Learning, Springer.
• Optional: David J.C. MacKay, Information Theory, Inference, and Learning Algorithms, Cambridge University Press. Freely available online.
• Optional: Trevor Hastie, Robert Tibshirani, and Jerome Friedman, The Elements of Statistical Learning, Springer. Freely available online.
• Optional: David Barber, Bayesian Reasoning and Machine Learning, Cambridge University Press. Freely available online.

Relevant Specialized Books (Optional)

• Carl Edward Rasmussen and Christopher K.I. Williams, Gaussian Processes for Machine Learning, MIT Press. Freely available online.
• Luc Devroye, Non-Uniform Random Variate Generation, Springer-Verlag. Freely available online.
• Daphne Koller and Nir Friedman, Probabilistic Graphical Models: Principles and Techniques, MIT Press.
• Jorge Nocedal and Stephen J. Wright, Numerical Optimization, Springer.
• Andrew Gelman, John B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. Bayesian Data Analysis, CRC.
• Thomas M. Cover and Joy A. Thomas, Elements of Information Theory, Wiley.
• Christian P. Robert and George Casella, Monte Carlo Statistical Methods, Springer.

• What is Metacademy? Metacademy is an exciting new online tool developed by Roger Grosse and Colorado Reed for helping you to develop personalized instruction. It's meant to help you manage what you know about different topics and develop an individualized curriculum to learn a new subject.
• I have an interview/sporting event/illness/computer crash. Can I have an extension? No. You can turn your assignment in up to a week late for half credit, and your lowest assignment grade will be dropped.
• Do I need to turn in code? No. You will write up your results with graphs, tables, and descriptions.
• How should I format my assignment? Use the provided .tex and .cls files (see the examples) to produce a LaTeX document. Compile it to PDF and upload the result to the iSites dropbox.
• What if I don't know how to use LaTeX? Everyone doing quantitative work should know how to use LaTeX, so consider this class an opportunity to learn it.
• Is attendance required at lecture/section? No, attendance is not required. However, you will be assessed on material that is presented in both lecture and section and may or may not be available from the readings alone.
• Can I do the final project without a partner? Yes.
• Can I have a group of three for the final project? Yes, but you'll be expected to do an amount of work appropriate for three people. I suggest discussing this with the instructors.
• I think I have found an error in the Murphy book! This is entirely possible. Let's share a comprehensive list of these in Piazza.
• What is your policy on simultaneous enrollment? You may not be simultaneously enrolled in this class and another.