A standard project report should take the style of a NeurIPS conference paper. However, this should not suggest that we expect publication-level results for a project to be a success. Your project should be attempt to accomplish your proposed research goals, and your report should reflect your work and results. If your intention is to turn your project into a submission for a conference, it will have likely already exceeded the expectations for assessment in the course.
If your project is not well suited to this rubric, for example if you are targeting a different publication venue, or are submitting an open source package, please feel free to propose alternative submission structure. This rubric should give a rough guide for our expectations, but off-rubric approaches are welcome. There are many ways to make contributions to our field!
Length of report should be 4 to 8 pages, not including appendices. Don't be afraid to keep the text short and to the point, and to include large illustrative figures.
Abstract that summarizes the main idea of the project and its contributions. Avoid jargon, make the abstract accessible to anyone in the course. Don't be exhaustive, don't list everything you did, just the main couple takeaways.
State the area and problem being investigated. Provide any general intuition, motivation, or background context. Emphasize connection to relevant areas of literature.
Show the overall model or idea.
High quality figures are valuable for making your paper more attractive and accessible to readers getting started or skimming. They should also help readers who are trying to understand a difficult aspect or crucial abstract concept to appreciate your work.
Figure should include a descriptive caption.
Describe your project formally by presenting the primary model / algorithm / loss function / conjecture in at least one of
- an algorithm box
- equations describing the model
- theorem or formally stated conjecture
Emphasize how your project is different from existing approaches, if appropriate.
If your work builds on previous work, clearly attribute the results to their sources. Include short paragraph summary of closely related papers and projects. Distinguish where your new contributions are.
Justify your idea with a demonstration, e.g. performance on toy data, or compare results against baseline models. Quantitative comparison can be a table of loss metrics. Qualitative comparisons are okay too!
Give an assessment of your experience with the methods you were investigating. Suggest settings where you expect the approach to perform well or poorly, especially if it contrasts with baselines or your initial expectations. Suggest future extensions, approaches to address limitations, or new open problems
Clearly state your achievements in the project and connect them to the main goals and ideas.