Assessment and workload:
– 20% Midterm test, no helpsheet (I got 14/50, which was absolutely depressing)
– 80% Final examination, 1 A4 sized helpsheet.
Experience:
– This module extremely difficult. It elaborates a lot about the concepts gone through in MA3252 for linear programming, and MA3236 nonlinear programming. Proofs that were previously omitted in those 2 modules are discussed here. As somebody who has done all optimisation modules, this module is a real eye opener for me because the concepts of optimisation go pretty deep. The module starts with a very thorough discussion about convexity, convex sets and functions, then moves on to optimality conditions. Chapter 3 would be about duality and the most general form of duality, Fenchel duality, would be introduced. The following 6 chapters would be optimisation algorithms and their convergence results. The final chapter would be about nonsmooth optimisation which is really difficult but interesting. This module does complete the optimisation package in NUS.
– Prof Zhao has taught this module many times and is very experienced in teaching it. He explains the concepts in the lecture notes pretty well and in a systematic way, adding his own understanding as input when he writes the concepts on the whiteboard. He is also very approachable and keen to answer questions posed by students.
– There are no tutorials for postgraduate modules but Prof Zhao provides a list of recommended exercises to try, which were pretty useless because his test and exam uses techniques of a way higher level compared to the exercises.
– Midterm test was a bloodbath. Prof Zhao set a pretty difficult paper, at least it was really difficult to me. The level of technique needed to do the questions is very high and it is not easy to come out with those techniques in the short span of 90 minutes. The cohort didn’t fare that well, and the mean is in the low 20s out of 60 before Prof Zhao voided 1 question. Still, my coursemates, especially the Phd students, are very strong and the highest scorer got 54/60 before Q5 was void.
Head over to our Shop for more module content!
