Due: Mon Oct 1 at 5:30pm. Weight: 10% (of final project grade)
You should form teams of 3 people for the course project. Smaller teams will
need to ask for special permission. Broadly, the writeup for the first milestone
should tell us what you are planning to do for the project. You have the option
of doing one of following three things:
Option 1 requires the most novelty, of course, but each of these options should
require roughly the same amount of effort if done properly, and we will grade
the projects accordingly. For example, we will generally expect more thorough
empirical comparisons under Option 2 versus Option 1. And if you choose Option
3, then the review should be thorough enough that someone new to the topic
should have a good understanding of the relevant works by the end of your paper.
- Choose a problem of some importance in an applied domain area, and design
a novel optimization method, with either an empirical or theoretical focus on
how it works.
- Choose a problem of some importance in an applied domain area, and
implement several existing optimization methods, showing how they work
practically (possibly in a new setting/regime that has not been considered
- Conduct a thorough literature review on a problem or method pertaining to
We strongly encourage students to put any implementations that come out of this
project up on Github, and made open-source.
Finally, regardless of which option you choose, your project should be
relevant to the topics covered in this course (e.g., don't just choose to
study deep learning unless you do something with it that relates to the convex
optimization tools learned in the course). For some inspiration, check out some
project ideas taken from a previous year
of this course.
The writeup for milestone 1 should meet the following criteria:
Submit your first milestone on Gradescope as a group submission by 5:30 pm
on Mon Oct 1.
- It should have the project title and andrew IDs of the team
- It should have a good introduction and background material,
typically covering 2-4 peer-reviewed papers. It should mention why
the problem is important and/or challenging. An abstract is not
- It should state as clearly as possible what you wish to accomplish
by the second milestone and by the end of the project.
- It should be limited to 2 pages, excluding references, and should
be completed in LaTeX in
NIPS 2018 style.
Grading breakdown: introduction (40%), clear problem statement (40%),