View All Pages

Projects

Final projects are worth 50% of your overall course grade: 2% for determining your project team and area; 8% for a 2-3 page project proposal; 15% for a short oral presentation; 25% for a final technical report.  Successful projects typically have the following key steps:

  • Find an application whose structure seems promising for probabilistic graphical models. Ensure data is easily available.
  • Identify a baseline application of machine learning (not necessarily graphical models) to related data. Your work will be easier if you can find existing research papers, technical reports, or software packages with relevant results.
  • Propose a graphical model, or a few related models with varying degrees of complexity. These models may be identical to ones discussed in lecture or the research literature, or minor variations motivated by your application.
  • Implement learning and/or inference algorithms that are appropriate for your graphical model and data. You are very welcome to build on existing software packages, but should do more than simply run off-the-shelf code.
  • Validate your approach via careful experiments. Application performance numbers are of interest, but you should also confirm the correctness of your learning algorithms, and inspect learned model parameters or structure.

Project expectations depend on the resources used, which your final report must acknowledge:

  • If you are reimplementing the learning/inference algorithms based solely on descriptions in research papers, then it is sufficient to validate by replicating the published results.
  • If you build on a public implementation of the learning/inference algorithm, then you should do more than run off-the-shelf code, and should extend/adapt the previous methods for your chosen application.

If you're interested in a different (perhaps more theoretical) style of project, talk to the instructor. PhD students are of course encouraged to identify topics which support their thesis research.

Project Teams (May 22)

You need to identify your project team, as well as proposed a general topic for your project.  We strongly encourage teams with two, three, or (at most) four members.  One way to form teams would be to use Ed Discussion Links to an external site..

If you would like to do a solo project, please email the instructor to request permission.  Typically students with sufficient background for solo projects have graduate research experience.

Identify your project team by submitting a brief description of your project area on gradescope Links to an external site..  One team member should upload the description, and then will have the option to identify the other team members.  (Project proposals, presentation slides, and reports will be submitted similarly.)  Use the "View or edit group" option on gradescope to be sure this is done correctly.  Do not complete the assignment multiple times; only one team member should submit.

Project Proposals (May 28)

The project proposal should be about 2-3 pages long, and no more than 4 pages long, including all figures. We encourage, but do not require, you to use the NeurIPS LaTeX style file Links to an external site. (use the "preprint" option so authors are visible). Proposals must be uploaded as a single pdf file. Your proposal should contain the following information:

  • A clear description of the problem or application you intend to address. Why is it worth studying?
  • A discussion of related work, including references to at least three relevant research articles or technical reports. Which aspects of your project are novel?
  • A figure illustrating a graphical model which plays a role in your project. We recommend creating such figures in a vector drawing program, such as Adobe Illustrator, Powerpoint, Keynote, or Inkscape Links to an external site..
  • A figure illustrating a preliminary experiment with some data related to your project. This could be some sort of visualization of the raw data, or the results of running a simple (supervised or unsupervised) machine learning method.
  • A description of the learning and/or inference challenges that you need to solve to apply a graphical model to your data. It is fine if you do not yet know what algorithms are appropriate, but discuss the challenges that need to be solved.
  • An experimental evaluation protocol. How will you know that you've succeeded?

Project Presentations (June 6)

Short project presentations will be given during the normal lecture time.  Your presentation should include:

  • Background material appropriate for an audience who knows the course material, but not your application area.
  • A clear, high-level description of the graphical model(s) and learning/inference algorithm(s) used in your work.
  • Visualization and analysis of your results. It is OK if you have not completed all experiments you intend to include in your project report, but there should be some partial results. Negative results (poor performance) is also OK, as long as you thoughtfully analyze why things aren't working as well as you'd hoped.
  • Contributions from all team members. We are evaluating clarity of presentation as well as the quality of the work, and thus each team member should participate in the talk.
  • About 8-10 minutes of total material for group projects.  Shorter talks (6-8 minutes) are fine for solo projects.

Project Reports (June 13)

The technical report should be between 6-10 pages long, in the style of top machine learning conferences. Although the results need not be sufficiently novel for publication, the presentation and experimental protocols should be of high quality.  We encourage, but do not require, you to use the NeurIPS LaTeX style file Links to an external site. (use the "preprint" option so authors are visible). Reports must be uploaded as a single pdf file. Your report should include:

  • A clear description of the problem addressed, and summary of related work with appropriate references. Include figures illustrating the graphical model(s) used in your project.
  • A mathematically precise description of the statistical models and learning algorithms that you consider. For the parts of your project which are novel contributions, include derivations which are sufficiently detailed for knowledgable experts to reproduce your work. Where you adapt previous work, included detailed references.
  • To help verify that your statistical learning algorithm is working properly, at least one plot showing the learning objective (joint log-probability for an MCMC method, a log-likelihood bound for a variational method, etc.) as a function of the number of learning iterations.
  • Some sort of visualization of the learned model structure; summary performance numbers are not sufficient. For example, for many graphical models it is possible to plot the learned clusters or features or states, sample from the posterior or predictive distributions, visualize results on low-dimensional toy data, etc.
  • A description of implementation details, including references for any code that was adapted and reused, a high-level summary of the functionality that your code implements, the programming language(s) you used, etc.
  • Mandatory: A description of how each team member contributed to the project.