Overview

The goal of the course project is to give students hands-on experience in a research area of their choice, related to the topics covered in class. Teams of 2 or 3 students are recommended. Solo projects are allowed with a justification (e.g., if the project is part of your ongoing research).

The following types of projects are welcome and encouraged:

  • Paper reproduction: Reproduce any paper discussed in class or included in the recommended readings, or related work from ML/NLP conferences. Your project should:
    • Include a subset of the original experiments, along with meaningful extensions (e.g., new datasets, base models, or ablations).
    • Provide novel insights or results beyond the original paper.
    • Meta-reproducibility studies across related papers are also encouraged.
  • Discussion-inspired projects: Develop ideas inspired by in-class discussion.
  • Ongoing research: You may build on your current research (including work with collaborators outside the class), as long as it clearly relates to the course topic and is noted in all deliverables. If unsure about relevance, consult the instructor.

Please keep in mind the compute resources available to you. The instructor can’t provide compute resources for the project.

Milestones and Deliverables

There are three main deliverables: project lighting talk, the project presentation, the final paper along with the code base.

Milestone 1: Project matching survey (Deadline: 09/16)

You are encouraged to build your team, feel free to use the slack channel to post your interest/ideas and connect with your classmates there.

Milestone 2: Project Lighting Talk (10/7)

8 minutes presentation followed by a 2 minitues discussion. The presentation will cover the motivation, key ideas, the concrete plan and the expected outcome of the project

Milestone 3: Project presentation (11/18, 11/20 and 11/25)

Each team will have a total of 12 minutes for their final project presentation, which includes 10 minutes for the presentation itself and 2 minutes for questions and answers. Every team member must participate and present a meaningful portion of the work. The presentation is an opportunity to communicate the core ideas, methods, implementation, and results of your project. While the written report will contain full technical details, the presentation should highlight the essential aspects that demonstrate your understanding, effort, and outcomes. Below is a suggested structure for the presentation:

Begin your presentation with a brief overview of the background and motivation for your project. Spend about a minute introducing the problem you aimed to solve and explaining why it is interesting or important. Provide just enough context for the audience to appreciate the motivation behind your approach, without going into extensive literature review or unnecessary detail.

Next, give a clear overview of your method. Describe your main idea or approach, the reasoning behind your design choices, and any important theoretical or conceptual elements. Focus on conveying the intuition and high-level flow of your method, rather than listing equations or code details. The audience should come away understanding what your system or algorithm does and why it makes sense for your problem.

Then, focus on implementation and effort. This is where you show what you actually built and how much work your team put into the project. Discuss the system architecture, tools, and techniques you used, as well as any challenges you encountered and how you addressed them.

After describing your implementation, move on to experiments and results. Present the most important findings that support your claims and demonstrate the effectiveness of your approach. Use figures, tables, or short visualizations to make your results clear and convincing. The results section should connect directly to your method and implementation, showing that your conclusions are grounded in evidence.

Conclude with key takeaways and reflections. Summarize what you learned from the project, what worked well, and what could be improved. Your takeaways should be supported by your experimental results and provide a clear sense of what insights your project contributes. Briefly mention any future directions or extensions that might be worth exploring.

In evaluating your presentation, we will consider clarity and organization, depth of understanding of the problem and method, quality of the implementation and level of effort, strength and presentation of results, and effectiveness of delivery. We are looking for presentations that communicate both understanding and accomplishment, showing that you not only learned something meaningful but also created something substantial.

Milestone 4: Final report and code repo (Deadline: 12/11)

  • Report:Use the unmodified COLM template. Find the latex template here. Up to 8 pages, not including references. You are welcome to include an Appendix with no page limit, but the evaluation will primarily be based on the main paper. We expect the report to generally follow the structure and style of a workshop/conference paper, including sections for the abstract, introduction, methodology, main evaluation and analysis, and related work. In addition, the report should include a section detailing each individual's contribution. If you're interested in submitting your work for real, feel free to reach out to the instructor for help.
  • Code: a link to a github repository containing your code and make it accessible to the instructor if it is private. If your repository is not visible to the instructor, your final submission will not be considered complete. We use this repository to check contributions of all team members.