Math 5750/6880 Optimization announcements

Grades

2008-12-27T08:16:43-07:00

Your grades for Math 5750 and 6880 have been submitted to the registrar. You should have received an email with your class grade, HW grade, project grade and presentation grade. Taking both classes together the average grade is a B+, and here is a histogram (with the two classes together):

A	########
A-	###
B+	##
B	#
B-
C+	##
C
C-	#
D+
D
D-
E	#

Reports

2008-12-09T16:51:42-07:00

Some comments on the Project Reports:

The official due date is Friday 12/12. However I will not grade them until Friday 12/19, so you have until 12/19 turn them in, on the condition that after 12/12 all reports should be submitted by email to the instructor. If you do not receive a confirmation email within 24h of your submission, please resend your report.
If submitting your report by email you can still write it by hand to save time. Ask Paula in the Math department office to show you how to use the scanner (only B&W scanner is available).
What I am expecting to see is about 5 pages (or more) on what is the problem you solved, what methods you used to solve it and a comparison of the different methods that summarizes the numerical experiments you ran.
After I am done grading the projects I will send you back an annotated PDF or an email with my observations and your grade for the project and the class.

More Project Typos

2008-12-01T21:13:42-07:00

Thanks to Hilary for spotting the following typos (corrected in the newest versions of project.pdf and eit.tgz):

File fobj_GNHessian.m, line 44 should be: H = J'*kron(eye(nb),pars.Qbdry)*J;
Fixed typos in Section 5 of project.pdf. In the description of the Gauss-Newton Hessian approximation computation (3) should be and (4): .

lecture20

2008-12-01T15:21:02-07:00

Today we started derivative free optimization methods. Lecture notes are available here: opti020.pdf (see also Chapter 9 in N&W).
Here are the Project presentation schedules. It is not necessary to bring your laptop for the presentations. Just bring your slides in a USB key or email them to me before class. Tou can also present with the slide projector or even on the board. The order of the talks below is not fixed.
- F 12/5: Seth (translation), Joey (radio tomography)
- M 12/8: Joshua (tumor treatment with heat), Tao (transportation), Hong (EIT)
- W 12/10: Jaggadeesh (semantic indexing), Jiarong (EIT), Hilary (EIT), Eric (financial), Andy (calibration)
- F 12/12: Shawn (optimal structure), Mingxing (EIT), Carlos (EIT), Qin (EIT), Scott (vascularization), Ge (EIT)

lecture19

2008-11-26T13:42:16-07:00

Today we continued with trust region method (including for SQP). Lecture notes are here: opti019.pdf.

Project Presentations

2008-11-24T21:10:52-07:00

The project presentations will be held on F 12/5, M 12/8, W 12/10 and F 12/12. Please email me if you have a preference, otherwise I will assign a day to you. You can present with slides, projector or even on the chalkboard.
- Common project: 5 min presentation. Describe your numerical experiments with the project problem and present your results.
- Other projects: 10 min presentation.
Common project tips:
- The gradient is easier to check if you set pars.V=eye(nbe); (or data for some conductivity) instead of the suggested randn(nbe).
- To check that the adjoint state method is correctly implemented (independently of the regularization terms) you can set pars.alpha=0. Regularization is needed to solve for the minimization, so do not leave pars.alpha as such.

lecture18

2008-11-21T16:37:01-07:00

Today we finished the discussion on reduced SQP methods. We started trust region methods, here are the lecture notes for today and Monday: opti018.pdf.
Fixed a typo in the project instructions project.pdf. The computation of the residual in section 4 ((b) in page 5) should be (Thanks to Hilary for spotting this).

lecture17

2008-11-19T14:55:52-07:00

Lecture notes for Wednesday are here: opti017.pdf, we covered reduced SQP methods.
A clarification on HW4: The convention for the sign of the Lagrange multipliers in the provided EQP solver is the one used throughout the book, namely:
. Your SQP algorithm should be consistent with this convention in order for it to converge.

hw4

2008-11-14T15:20:23-07:00

Homework 4 is due Monday December 1st 2008 and is here: hw4.pdf. Supporting code is:

eqp.m An equality constrained quadratic programming solver using the nullspace approach
test_eqp.m tests the above code on a the simple Example 16.2 of the textbook.
f_18_3.m The objective function in Problem 18.3 in the textbook.

Project Update

2008-11-14T09:30:00-07:00

The common project support files and instructions have been updated: project.pdf and eit.tgz. A demo will be given tomorrow.