The course structure will closely follow the format of Colin Raffel’s COMP790 course at UNC. (See also Colin’s original blog post proposing this format). The course will consist of two components: paper readings and a final project.
Paper Readings
Most classes in the semester will involve the discussion of a single paper. Everyone is required to read the paper before the class. A subset of the students will be assigned “presentation roles” on a rotating basis for which they will need to come prepared. Other students will need to complete a short exercise before the class and come prepared to discuss. The presentation roles are designed to expose students to the different aspects of being a researcher. By the end of the course, students should be comfortable reading, reviewing, implementing and extending NLP papers.
Presentation Roles
The following describes the responsibilities of the students assigned to one of the four roles we will use in this class. Multiple students may be assigned the same role in a class and they need to work independently.
- Reviewer: Must complete a full conference-level review of the paper. Some guidelines are available at the NeurIPS website (under Review Content). You will need to submit a write-up of the review before class (answering questions 1-10) and be prepared to discuss it in class.
- Hacker: Must implement a small part of the paper on a small portion of the data (or toy data). We encourage the use of Google Colab for writing and running the code. You will need to submit the code and be prepared to share it in class, discussing any challenges you faced. Note that you need to do more than just download the authors’ code and run it — you should implement at least one core part of the paper.
- Archaeologist: Do a thorough investigation of the prior work in the context of which this paper sits. Also, look for newer papers which have been influenced by and cite the paper being discussed. You will need to make a few slides and present them in class.
- Researcher: Think about any ethical and social considerations arising from the paper and propose an imaginary follow-up project to the existing paper. The follow-up should be something which is enabled by the research discussed in the paper. You will need to make a few slides and present them in class.
Non-Presenter Exercise
Students not assigned one of the roles above should come to class prepared with:
- An alternative title for the paper, or a new name for the algorithm it proposes.
- At least one question about one of the experiments in the paper, or an additional experiment missing from the paper.
Final Project
All students in the class also need to complete a final project in teams of 2-4. The project must extend one or more papers discussed in the class in a novel way. Students are expected to write code for the extension, run experiments and report on their findings in a standard conference format. Larger teams are expected to put in more effort. More details coming soon!
Grading
The final grade will consist of:
- Background quiz: 5 points (submission is enough to get full credit)
- Paper Readings: 60 points. (details coming soon)
- Final Project: 35 points.
