Discussion and project resources

Paper discussion format

The course will have 15 student groups and 15 paper-discussion classes. Groups should aim for three students whenever possible, and each group will serve exactly once in each role: presenter, advocate, and critic.

For each discussion, there will be one required paper from the reading list. The presenter group should also choose one additional paper from the related readings so that the class discussion can compare two closely connected ideas rather than treating the required paper in isolation.

The presenter group is responsible for a 35-minute presentation that explains the problem setting, background, current practice, key insight, technical approach, evaluation methodology, and main results in a way that is accessible to the full class.

Presenter group guidance

The presenter group should structure the talk around the following questions:

1. What problem is the paper trying to solve? State the objective clearly and with as little jargon as possible.
2. How is this problem addressed today, and what are the limits of current practice? Situate the paper relative to the state of the art.
3. How does the paper motivate its key ideas, and what are those ideas?
4. What is genuinely new in the proposed approach, and what evidence does the paper provide that the approach is effective?
5. How are the ideas evaluated? Explain the evaluation methodology, the metrics, and the paper's most important empirical results.
6. Why does this work matter? If the paper's claims hold, what meaningful difference does it make for the field or for practice?

Advocate and critic group guidance

The advocate and critic groups are responsible for leading a substantive debate on the strengths and weaknesses of the paper.

The advocate group should begin by identifying three major strengths of the paper in a concise and well-supported manner. The critic group should begin by identifying three major weaknesses, limitations, or unanswered questions, again in a concise and well-supported manner.

These opening points should serve as the starting point for a broader discussion with the rest of the class. In all cases, the goal is to be analytically rigorous and intellectually fair. Students should focus on substance, evidence, and reasoning rather than rhetoric or personal language.

Before the broader debate begins, the class will have an opportunity to submit comments and criticism through a Google form. The critic group may use those audience submissions to sharpen its case, and the instructor will moderate the debate.

Class timeline and logistics

A typical discussion class will follow this structure: 5 minutes of context and setup by the instructor, 35 minutes for the presenter group, 5 minutes for the class to discuss and submit comments, 5 minutes for the advocate and critic groups to review those comments, and 20 minutes of moderated debate.

The audience is expected to read both papers before class, ask questions, submit criticism after the presentation, and participate actively in the discussion. Audience participation is part of the course grade.

Groups should first sign up for teams, then sign up once each for presenting, advocating, and critiquing. The presenter group must upload slides to Canvas at least 3 days in advance: by Saturday at 11:59 PM for Tuesday presentations, and by Monday at 11:59 PM for Thursday presentations.

Evaluation metrics

The instructor will evaluate participation in paper discussions using the following criteria:

• Presenter group: quality of presentation; participation from all group members; clarity of communication; accuracy of the paper summary and depth of understanding; ability to answer questions; and ability to complete the presentation on time.
• Advocate group: ability to identify and communicate the paper's three most important strengths succinctly; ability to defend those strengths during discussion; and ability to argue in a substantive, respectful manner.
• Critic group: ability to identify and communicate the paper's three most important weaknesses succinctly; ability to defend those critiques during discussion; and ability to argue in a substantive, respectful manner.
• Audience: quality of participation in the discussion, including the ability to raise important points, ask thoughtful questions, and contribute constructively to the conversation.

Guidelines

Students may find the following resources useful while preparing for paper discussions and debates:

• Use this presentation resource to learn how to give a good presentation.
• Use this paper-reading guide for advice on how to read research papers efficiently and critically.
• The advocate and critic groups should consult this guide to writing systems conference reviews when preparing their arguments.

Debate should remain rigorous and civil. Do not attack the authors, do not make personal comments about other debaters, and do not confuse disagreement with disrespect. Focus on substantive arguments, evidence, assumptions, limitations, and possible improvements.

Choosing a project

Projects should focus on a concrete systems or hardware problem in AI rather than simply building an application or reproducing an existing result. Strong projects are organized around a clear research question and a realistic plan for investigating it.

Defining success

Every project should identify at least one primary quantitative metric, such as throughput, latency, utilization, memory footprint, energy, or cost. Teams should also define baselines and report secondary metrics when they help explain tradeoffs.

Proposal and milestones

Teams should submit a proposal early in the semester that defines the problem, research question, metric, approach, evaluation plan, feasibility, and risks. By the mid-semester review, each team should already have a working baseline and initial characterization results.

Final deliverables

The final project should tell a clear research story: problem, observation, design, evaluation, and conclusion. Teams will present their strongest result at the end of the semester and submit a concise final report supported by rigorous measurements.