My Review of (ISYE 6669) Deterministic Optimization

           Grade: A
      Difficulty: 7/10
          Rating: 9/10
 Time commitment: 9 hours/week
-------------------------------

Overall

This course covers various optimization topics with a special focus on convex optimization and linear programming. The first half of the course focuses on theory/math (linear algebra review, set theory, convexity, epigraph, geometric objects like polyhedron & polytope). The second half focuses on algorithms to solve optimization problems.

Lecture Quality

• I found the lecture video quality to be excellent. Very well organized.
• But they move fast and cover many topics. The head TA does warn on the first day of the class "this is a fast paced class."
• Also, there are a few typos in the lecture slides, so you have to watch out for the TA announcement to be aware.

Assignments

• Weekly homework (20%) :
  • There are 15 weekly homework assignments. It's not hard but takes a decent amount of time. Took me anywhere between a few hours to 10 hours. Once in a while, there is a question that requires Python coding but usually less than 20 lines of code.
• Midterm exam (40%) :
  • 30 questions. Multiple choice format. 180 minutes, proctored and closed book, but they allow one sheet of paper with hand written notes (aka cheat sheet). No calculator is allowed. A few questions require calculation (matrix rank, partial derivatives, hessian matrix, simplex method, so on) but they are reasonably solvable by hand. Overall the midterm is not hard, and closely follows the format of the "practice exam." Although they allow 180 minutes, each question typically only takes less than 3 minutes. For reference, I finished the whole midterm in less than 90 minutes. No coding is involved.
• Final exam (40%) :
  • Same format as the midterm exam. 30 multiple choice questions. 180 minutes, proctored and closed book. I felt the final was slightly easier. There are no "bonus" questions. All questions are real questions that require solid understanding of the materials. But also there are no trick questions, so the exam is fair.

Grading

• HW is peer reviewed. You get assigned the median score out of 3 anonymous peer reviews. I usually got 90 ~ 100%. They drop the lowest two homework scores.
• Midterm/final : The course average exam score was approx. 81%. The median was approx. 83%.
• Overall, no grading curve. The 90% is the hard cutoff threshold. Approx. 25% of the class got an A.
• The biggest pressure of the grading is, to achieve an A (90%), you can only get at most 6 (out of 60) exam questions wrong, otherwise you fall below the 90% threshold. I managed to get 56/60 (= 93%). But if I missed 3 more questions wrong, then my grade would've fallen below 90%.

Thoughts

• It's become my favorite class in the whole OMSA program. I found it super fun. You should definitely take this class if you are interested in the subject.
• It's a pencil & paper math course. Mostly theory, with very little coding involved. Just a few HW questions get you to write python cvxpy code, but very minimal.
• Yes, the fact that two exams (total of 60 questions) mostly determine your course grade is super stressful. I wish they give more weights to the weekly homework which takes considerable amount of effort to complete.
• This has been the feedback by many students for many years but the instructor has not changed it, so it will stay this way.
• If I may creatively brainstorm ideas on how to improve this class, I'd suggest they introduce a mini project instead of a final exam, and let students work on well defined optimization problem with enough complexity, like stock portfolio optimization.
• But knowing the instructor's personality (a very much old school math teacher), I'm convinced this course will stay with the current "exam decides your grade" format.

Course administration

• Like many other OMSA (and OMSCS) courses, the professor on record (Santanu Dey) is completely absent, so the TAs are running the show. I thought they were all well oreganized and dedicated. They answer questions on Piazza quickly and thoroughly. I could tell they really love the subject.
• I attended a few office hours by different TAs, and found the head TA to be the most knowledgeable and her office hours were super useful. She does have quite a bit of Eastern European accent and her words may come off as blunt but she really cares about teaching and helping you learn.

Resource

• If you take this course during Spring semester, you can get free tutoring service from students who are verified to have earned letter grade A in the course in the past semesters. (GT pays the tutors, so it's free to students). It's done on a platform called Knack.
  https://grad.gatech.edu/knack-tutoring
  • I got immense help from my tutor. Definitely couldn't have achieved an A without his help. Somehow they only make this course available on Knack during Spring semester.
• Feel free to ping me. I can probably give useful tips.

FAQ

• [Prerequisite] How much math do I need to review & master before taking this course ?
  • My undergrad background is CS, and I don't consider myself good at math. But honestly, to my positive surprise, the math in this course was very manageable. Mostly basic linear algebra, with a bit of multivariate calculus and set theory. They have a quick math review module in the 2nd week, so I didn't really need to review anything before the semester started.
• How hard is it to achieve letter grade A in this course ?
  • I investigated this topic. Because there is no grading curve, to get an A means to achieve at least 90% score.
  • First, it's nearly impossible to score 100% average on weekly homework assignment, partially because you will inevitably make tiny human error occasionally, and partially because some peer reviewers don't want to give you 100% so they just give you 90% for no reason. Let's say you got 95% average across 15 HW assignments.
  • Now, let's consider how many questions you can afford losing in the midterm/final. Each exam question is worth 1.33% of the overall course grade. As below, out of 60 questions in total, you can only get up to 6 wrong. Or alternatively, if you can somehow manage 97% average score on HW, you can afford to lose up to 7 questions on the exam.
  • At any rate, there is very little margin of error allowed before your grade drops from A to B.

Reference

• Syllabus : https://omscs.gatech.edu/isye-6669-deterministic-optimization