CPSC 532B

 Announcement  Location updated to ORCH 4074

Syllabus and Course Information

Welcome to CSPC 532B, taught by Peter West. We will learn about open questions in NLP and AI, while providing the necessary background for new researchers to start tackling these problems. Where should someone who wants to research GPT-4 GPT-5 GPT-6 start?

This will be a seminar course, with paper presentations and a course project being the primary coursework. We will develop core academic skills: understanding research papers, their context, and their motivation; presenting research in a way that is clear and compelling for broad audiences; useful feedback and discussion; basic research project design and work.

Students will be graded based on:

(1) Paper Presentation. [40%] These will be solo or in pairs, depending on available slots. You should expect to present papers for roughly 40 to 50 minutes and lead discussion for the remainder of the class period. Submit your slides by email after your presentation, along with an explanation (max 1 page) of the decisions you made in designing the presentation and discussion. For pair presentations, both students should submit their own explanations. See more information below.

(2) Course Project [40%] These will be completed in groups of 2 or 3. Project proposals will be submitted in early October (due date: TBD). The basic requirement for the project is to ask a directed question that can be answered with a small number of scoped experiments. We will provide each group with $20 of credit on OpenRouter, on request. It is also alright to complete a project that connects to your own research, or utilizes other resources, as long as it is somewhat related to the topic of the course. Feel free to reach out to the instructor or TA if you have questions about project scope and topics. 

(3) Active participation. [20%] This is largely based on participation, so take part as much as you can! Although you don’t need to be an expert on every paper that is presented, you should read every paper in advance to make sure you are familiar enough to discuss it. Discussion includes asking (respectful) questions anytime you are confused--this helps you learn material, and presenters learn about their audience. Given the large class size, being an active participant in every discussion may be difficult, but you should aim to attend most lectures (at least 80%) and regularly take part in discussions/ask questions. 

As part of "active participation," all students should also sign up to give feedback on at least 3 presentations. Sign up for feedback slots on the course schedule. Feedback should consist of: one element of the presentation you thought worked well, and one thing you would add to the presentation to make it even stronger. These can be short (one sentence) but should be thoughtful and specific to the presentation.

Office Hours and Communication

Peter: Available after class most Tuesdays and some Thursdays, starting in the classroom and continuing in his office at ICICS 109 (NOT X-wing). He will also be available to meet by appointment to discuss presentations, projects, and course logistics. Feel free to email with any short questions or concerns. Please include "CPSC 532B:" in the subject line to ensure a timely response!

Alex: TBD

Presentation instructions and advice

Presentations should take roughly 35 to 50 minutes, summarizing the given paper and its significance. The two main goals are teaching your classmates and getting experience presenting. You have a lot of freedom in how you structure your talk--there are many ways to make a great presentation (e.g., focusing on critique, future ideas and open questions, or historical background). See this slide deck for some key requirements and suggestions on how you can format your talk.

Remember to submit your slides and a maximum 1-page explanation of how you designed your presentation (by email). Pair presenters should both submit their own explanations. These can be short and should simply provide an explanation of any challenges and design decisions that arose in making a clear and effective presentation, as well as understanding the paper. This is where you can provide some extra context if your presentation did not go exactly how you wanted.

I am always happy to meet with students before they present to provide feedback and guidance, and/or answer questions by email.

Project instructions and advice

Description

The goal of this project is to give you the opportunity to collaboratively study an open question in NLP, possibly building on the papers presented in the course. Some possibilities include:

• Extending work presented in class using new analysis or additional methods

• Applying methods or analysis from NLP to a novel problem in another domain (possibly your domain of expertise)

• Doing an initial study of a new problem or idea that you propose

Feel free to be creative here! You will submit an initial proposal for your team’s project by October 11, and we will provide feedback to help you scope the project to something meaningful but feasible. You can also contact the instructor or TA before the proposal deadline if you would like early feedback, and/or attend the project office hours on Zoom during the regular class time on October 8.

Team

The project should be done in teams of 3-4 students. In the final report, you will have the opportunity to list the contributions of each team member. Teams should be added to the “Project Teams” tab of the course schedule before the proposal deadline on October 11. You can form teams in class, or add yourself to a new or existing team, noting that final teams must have at least 3 students. 

Resources

We will provide each team with a limited-credit OpenRouter API key upon request (email the instructor with your team number). This provides inference-only access to a range of models. Please read the OpenRouter API documentation for information on how to track your use to properly budget. The key must be stored locally and should not be made available on GitHub. You may use your own resources (e.g., lab servers) if you plan to do a project that goes beyond LLM inference, but this is not required -- inference-only is sufficient.

As a team, you should make sure your project is realistic considering the resources available to you. 

Project proposal (due October 11th, on Canvas or by email)

Each team should write a short project proposal (aim for 2-3 paragraphs, max 1 page). The proposal should clearly articulate the research questions and how they will be approached by the team. Most importantly, you should describe at least one key experiment you plan to run. The proposed project should be simple and viable considering resources and time limits. The proposal can include some “nice-to-have” elements, if time allows. The proposal should also include references to any relevant literature that your project will build on.

The goal of the proposal is to obtain feedback before starting the project, but the project can evolve and change throughout the semester. You can always reach out to the instructor or TA for additional advice. Although the proposal will not be directly graded, it is mandatory, and required for full marks on the final project.

We will be hosting office hours on Thursday, October 8th during the regular class time over Zoom, to provide feedback on project ideas before the project proposal is due on October 11th. Your proposal does not need to be perfect, and we will provide feedback after it is submitted. 

Final report (Due date TBA, late November)

The final report should be 4-6 pages excluding references, following the ACL template. The report will be evaluated as a whole, you are responsible to organize it in a way that best presents your research. It must contain:

• Abstract. Summarizes the main idea of the course project and its contributions or findings. 

• Introduction. States the problem being solved and why we want to solve it.

• Core method and results. This will depend on your project, but should contain about your methods, experiments, and results.

• Related work. Includes closely related papers, while motivating how your contributions differ from them.

• Conclusion and limitations. Here, you can mention any parts of the proposal that you were unable to achieve and why.

• List of contributions. You should list the specific contributions of each team member.

Presentation

We plan to hold 1 to 2 mini poster sessions during the last few class times. More details to come!

Todos for each student

(1) Sign up for 1 presentation slot below (up to 2 presenters per slot, but all slots should be filled)

Please sign up by September 16

(2) Prepare your presentation, present, and lead the discussion

(3) Submit your slides and explanation by email

(4) Take part in discussions

(5) Provide feedback for other presentations through Canvas (at least 3, sign up in the schedule)

(6) Prepare your project (information TBA)

Have fun!

Expectations

While academic discussions can often be lively and filled with debate, all students are expected to treat each other with respect and compassion. Students will have varying levels of experience and confidence, so we should all strive to make this a comfortable space for everybody to explore new ideas. Research can be supportive and fun if we decide to make it that way. 

Bullying, academic dishonesty, and any form of discrimination or harassment will not be tolerated. If you have any concerns, please immediately contact the course instructor or department. 

We allow the use of generative AI to help with writing and designing presentations. However, we expect core content to come from students rather than AI, and hold students responsible for the content of their slides and writing, regardless of whether a mistake or idea was introduced by AI. Remember that a large aspect of this course is presenting and taking part in discussions, so simply submitting strong slides/writing using AI will not be enough to succeed in the course, and including generated content that you do not understand may result in questions that you cannot answer. As well, this course is a great opportunity to develop your own research, presentation, and writing style, which over-reliance on AI will get in the way of!