Sep. 15     There will be two textbook on reserve in the Taylor library. One should be available by tomorrow and should be a 2 hour reserve. The other should be available in about a week and will be on a 1 day reserve.

Sep. 15     Error Data for the book is available form the textbook website: Website for the textbook

Sep. 16     Problem 2 in assignment 1 needed some slight corrections, the new version is posted now.

Sep. 22     It has come to my attention that we only have a limited number of licenses for Matlab. Thus if have finished using Matlab, please exit the Matlab program so that other people can use it. Also the gaul (undergraduate) network has more licenses than the graduate network. If you are a graduate student and can't run matlab from a graduate lab, try going to the undergraduate lab.

Sep. 27     Lecture 3 slides are now corrected. Instead of 6.63 for the case of ML classifier it should have been 6.70. This correction should be made to slides 28, 30, 34, 53. Your slide numbers may differ from mine, if you do not have the most recent version of the lecture.

Sep. 27     Office hours this Wednesday, Sep. 29 will be held in room 300. Please come to see me if you have any questions about homework, which is due on Monday, Oct. 4.

Sep. 27     Correction to problem 1: Correction to problem 1: I forgot to mention that when you plot the histogram, the value at each bin should be divided by the bin width (in addition to dividing by the number of samples). If you do not divide by the bin width, you are plotting the probability mass function for the distribution of samples. To get the approximate density function for these samples, you need to divide by the width of each bin. Why you also need to divide by the width of each been will become clear after the lecture on Monday. Since I have discovered this bug rather late, if you are done with the homework and do not wish to correct it, you can hand in your solution as is without any penalty.

Oct. 7     Solutions and matlab code for Assignment 1 have been posted under October 6 in the schedule.

Oct. 12     Assignment 1 due date is extended until Wed., October 20, and also I won't apply late penaly if you hand it in by 4pm Friday, October 22. You should try to finish it by the 20th though because there will be a new assignment given out on the 20th.

Oct. 12     I have to move my office hours on Wed. Oct. 13. to 2:30-4:30.

Oct. 12     It is time to start thinking about the final project .

Oct. 25     I've changed the assignment and final project schedule. Assignment 3 is now due on Nov. 10, Assignment 4 is due on December 1, and there will be no Assignment 5. The schedule for the final project is as follows:

• Final project proposal is due on November 1. The late penaly is 1 point off the total score for final project for each day late.
• Data must be ported into matlab by November 15. Email me the .mat file which holds the data in matlab arrays. Also include a very brief description of what the data is and how it is stored, including number of samples in the training/testing sets and number of features. The late penaly is 1 point off the total score for final project for each day late.
• Make an appointment the week of Nov. 22-26 to discuss how the final project is going. If you prefer, you can give me a partial final report instead of meeting. If you do not with me, or meet with me but I see that you have done nothing at all yet, 5 points off the score for final project.
• Final project is due December 8.

Oct. 28     I had to make some changes to problem 2 of assignment 3 because generalized eigenvalue solver does not work well in Matlab.

Nov. 1     Part (e) and (f) of problem 1 in assignment 3 has been clarified.

Nov. 3     For assignment 3, make sure you use [V,D] = eigs(). If you just say V=eigs, this function returns the eigenvalues in column vector V. If you use [V,D] = eigs(), then it returns eigenvectors as columns of V and eigenvalues on the diagonal in D.

Nov. 3     I have corrected the "non-sample" midterm and solutions, and I will put them in the envelope on my door. Nov. 16     Teaching evaluations will be conducted on Dec. 1 at the end of the class

Nov. 29     Correction to lecture 13, slide 24. For the "true" gradient descent, the input-to-hidden weights w's should be updated first, followed by update to hidden-to-output weights v.

Dec. 8     Unfortunately I lost all the data for your final projects, so you need to resubmit it again. You can email it to me, or if it's large, put it somewhere on the web for me to download, or on a CD-disk.

Dec. 16     The final grades are posted