课程内容:四大门派。第一部分无约束优化,梯度下降、牛顿法还有它的变体全体出列;第二部分KKT条件、线性规划、单纯形法;第三部分对偶理论;第四部分凸优化,投影梯度、交替方向乘子法、不可行起始牛顿法、内点法一个不少。每节课发讲义,英文的,讲的全是讲义上的东西。参考书传承上Numerical Optimization和Convex Optimization都有。
上课自由度:签了两三次到,全是纸质签到,古法打卡。
考核标准(懒得写了直接粘贴):
• Homework (30%). If coding is needed, you can use whatever language you prefer. Python and MATLAB may be easier to get started. Please indicate your designed algorithm in the homework and show your results via figures or tables. The actual code should be submitted as appendices only.
• Quiz (15 %) There will be three-four quizzes over the semester. In each quiz, there will be 1 problem and you solve them in 10-20 minutes.
• Independence study (15%). You choose one topic from the following three which we do not have enough time to cover in class: 1) trust region methods; 2) conjugate gradient methods 3) quasi-Newton methods (in particular, Chapter 4, 5 and 6 in the book “Numerical Optimization”) by yourself. Use what you studied to solve an application you found.
• Final exam (40%) There will be a final exam. One cheating sheet of size A4 is allowed.
授课质量:较一般,这个授课质量建议自学,我基本都是辅以看书或者看讲义自学。老师基本上是对着讲义抄,每节课都给人一种备课极其不充分的感觉,就那个单纯形表法的例子,老师上课自己都没算明白,每节课都要卡壳几次。当然可能是老师科研确实比较忙。助教批作业极其严格,扣分扣得很狠,只看正确率不看做的认真程度,有点离谱。不过作业答案助教都做得很认真。小测题目还好,肯定没期末考题难。期末考试非常难,只能等老师捞了。综上这门课内容挺丰富的,去看那两本教辅一定会大有收获,单单靠课上听老师讲感觉学到的很有限,而且这个授课质量真的很容易听不下去。
如果想了解这门课作业难度可以看看[2024年秋6次作业及答案](https://share.dyweb.sjtu.cn/course/18290)