Final Project Description

We will be doing a final project instead of a final exam. The goal of this project is to gain a better understanding of the concepts we learned in this class, as well as to practice communicating about math and statistics. You can work individually or in groups of up to three people total. If you have multiple people, your final result should be more comprehensive than if you’re working alone.

There are three main components that will be evaluated:

  1. project proposal (10%)
  2. project presentation / review of your peers’ projects (40%)
  3. written final report (50%)

Each aspect will be given a rating “does not meet expectations”, “satisfactory”, “good”, and “excellent”. These will roughly correspond to letter grades of F, C, B, and A respectively. Based on your project proposal, I will provide you an estimated rating for your final report, assuming that you complete everything outlined in the proposal.

There are several possibilities for the contents of the project. You should pick one of the following:

Proposal (due 11/22)

The first part of the final project is a project proposal, and each group should turn in a single proposal.

The proposal serves several purposes. First, it forces you to start thinking about the project earlier than the last minute. Second, it gives me a chance to make sure your project is on the right track to satisfy the requirements. Finally, it will give you an outline to provide structure for the rest of the project.

Since the exact format for the project is very flexible, the project proposals are very important in making sure that your proposed project will be sufficient. Submit to Gradescope 1-2 pages outlining your plan for the project. Make sure to answer all of the following:

Presentations (12/13 in class)

Each group will prepare a poster on their project. The poster in not meant to be the same as the final report. Rather, the goal of your poster is to to explain what you did for your project to your classmates. It should not be a wall of text which is very hard to read.

Try to think about how to simplify complex mathematical ideas into something which is understandable to people with less background in the area. Remember, you’ve spent more time thinking about your project than anyone else has, so some things which you might think are ``obvious’’ may actually be something people might have questions about.

Posters should be modeled after ``research posters’’. NYU has some general guidelines here: .

Minimum requirements:

Blank poster boards are available for several dollars at most convenience stores or art shops. You can then print text onto standard letter paper and attach it to the board or write directly on the board. You’re not expected to have your poster printed by a poster shop, as this can be fairly expensive.

On the day of presentations, you will alternate between standing with your poster to present it and going to see other groups’ posters. You will write a brief review on several of the projects which you visited, and a form to assist with this will be provided to you on the day of the presentations.

Written report (due 12/21)

In addition to a poster, each group will produce a written report summarizing what you did during your project. The exact requirements for the report are flexible since there are many possible projects. Thus, you should lay out a precise plan for what you will put in the written report in your project proposal. Overall, the project is exepcted to take the time of several homeworks as it carries 30% of the final grade.

I have listed minimum expectations which you can use as a starting point for your proposal. However, your proposal should contain a detailed description of your plan for the project so I can determine whether it is sufficient.

Project Ideas

Below are a collection of resources you can use to help pick a project.

Datasets

Textbook’’ topics

Research topics

\begin{itemize} - Reproducibility in Learning - Reproducibility is a cornerstone of science, but many studied and experiments fail to be reproducible. This paper introduces the idea of reproducible algorithms, and describes how they can be implemented for many problems in learning theory. - A reproducible learning algorithm is resilient to variations in its samples — with high probability, it returns the exact same output when run on two samples from the same underlying distribution. - The paper covers a lot of different classes of algorithms, but the simplest example is statistical queries (section 2) which covers a number of the basic parameter estimators we’ve seen in the class (sample mean, sample variance, etc.)