Project

Overview

The final research project is meant to be a semester-long, very significant piece of work that demonstrates your developed understanding of privacy in machine learning. An ideal scenario is that your project becomes publishable in top-tier machine learning conferences. There will be a number of project ideas that will be given to you, in case you have a hard time finding one. Most of the projects would require you to write code to test your ideas, derive your algorithms, and mathematically prove that your algorithm is indeed differentially private. If you are really lost and do not know what to do for a project, you can do a literature survey on a subtopic, while this is generally not recommended.

Successful projects are those that produce a true publication-worthy work. As Frank Wood does in his grad course, I would also give high marks for projects that are ambitious and interesting than safe projects that produce an expected but positive result.

I recommend people to work in a group, consisting of 2 people, strongly preferred. A project team may consist of up to 3 people maximum. However, such a large group needs to do more than double the work of a two-person group. You can also work alone if that's what you prefer.

One of three different kinds of projects can be undertaken:

  • Programming a new or expanding the capabilities of an existing method/algorithm.

  • Developing a novel application using an existing algorithm with making modifications suitable for the particular application.

  • Making a fundamentally new DP algorithm or methodological development.

Do not wait until the middle of the term to start your project. As every good work requires, you need to pour many hours and thoughts into the project to be complete and successful.

Deliverables

Schedule [The description below follows the same rules as those in Frank's grad course]

  • Proposal due : See Syllabus (Please upload it to gradescope).

    • The proposal will be graded and will count as 20% of the final project grade. The proposal should take the form of a research abstract, i.e. it should include a title, authors, abstract, introduction, background/related work, suggested methodology, expected conclusions, and references. Note that the only significant missing pieces should be the actual methods and results. It should be extremely clear what the problem being addressed is. “Stand-in” figures that show what the expected results will look like are required.

    • All sections will grow in the final project report, but those listed should appear in the project proposal in reasonably complete form. Of particular interest are a description of the problem to be solved, a literature search, and expected conclusions to be drawn from a successful project.

    • Project proposals that do not outline a project with sufficiently high complexity will be given zero credit and groups that author such proposals will be required to resubmit project proposals on a weekly basis thereafter until the proposal reaches a sufficient level of complexity to warrant a passing grade. Resubmissions will receive strictly lower grades than proposals that pass the first time.

  • Assigned final exam slot : Final papers due and final project presentations are due at this to be determined time.

    • The presentation will be graded and will count as 10% of the final project grade. Each group will present their results via a presentation in front of the whole class. The duration of each presentation will depend on the total number of groups formed, but almost certainly will not exceed 20 minutes.

    • A paper copy of the final project report must be delivered to the instructor’s departmental mail slot by 12:00am, midnight, at the beginning of the day scheduled for final project presentations. Failure to submit a physical copy of the final project report by then will result in a grade of "?" for the final project.

    • Slides for the presentations (plan on 1-2 per minute of presentation) must be delivered electronically to the instructor via the class email by 12:00am, midnight, at the beginning of the day scheduled for final project presentations. The presentation should introduce the problem, highlight the methods used, and cover analysis results. Each group will be subject to 5-10 minutes of questioning from the instructor, potentially the other students in the class about all aspects of their project. Having backup slides (slides prepared, but not used in the presentation) for questions about methodology detail will be expected of all groups.