Presentation

Overview

At the beginning of the term, you will be given a chance to pick a paper you will present during the term. There are roughly four categories of papers from which you can choose:

  • Popular Differentially Private Machine Learning Algorithms

  • Differentially Private Data Generation

  • Differential privacy and other notions (causality, fairness, and interoperability) in ML

  • Privacy attack methods

The paper list can be found in the Syllabus. In case there are more students than the number of lecture slots to present, we can increase the number of papers to present in each lecture, which makes a team of presenters for each lecture.

Preparation

The presentation will be 60-65 minutes long, followed by Q&A and discussions for 10-15 mins. In your presentation, try to focus on these points:

  • What problem is the paper trying to solve?

  • Why do we care about this particular problem?

  • What are the SOTA (state of the art) methods?

  • How does this paper improve the SOTA?

  • How does the method work? If necessary, spend 20 mins or so to go over proofs if the theory part of the paper is what's novel.

  • What are the key results?

  • How is it evaluated?

  • What are the pros and cons of the proposed method?

  • What related problems are still open?

  • Why do you like/hate the proposed method?

Preparation meeting with the instructor

You're required to meet with me a few days before your presentation (Please email me to arrange it, as there are no set office hours for this course). Before this meeting, read your assigned paper very thoroughly, and prepare some or full slides if you have time. During the meeting, we can discuss details if anything in the paper is unclear. You can do a quick test run with me if you already have slides fully prepared, or request whatever way I could help you with the presentation. I wouldn't emphasize the importance of eloquence in the presentation (although it would be nice to have that). But what's important is clarity in the key points addressed above. What matters is whether the students in the classroom, including yourself, feel like they learned something useful from your presentation.

Peer-to-peer grading

At the end of each class, I will send out a Google form, in which you can put the grade of the presentation and also put your feedback anonymously. I will then share the result with the presenter so the presenter can also use the feedback to improve his/her/their presentation skills in the future. The grade is between 0 and 10, where higher is better. For grading the presenter, please consider the following aspects:

  • Did I learn something useful today?

  • Was I able to follow the presentation?

  • Was my question well answered by the presenter?

  • Did the presentation cover several of the points above (written under Preparation)?

After the presentation

Email the instructor the presentation slides, which will be posted in the Syllabus.