# Project Presentations **Dates: April 28–30 (schedule TBD)** The presentation is your opportunity to share what you've built and learned with the class. This isn't a polished product demo — it's a research talk. Your audience wants to understand your question, your approach, and what you found out, including the complications. Talks will be 20 minutes followed by 5 minutes of questions. That's a tight window, so practice and time yourself beforehand. --- ## Structure Your talk should cover the following, roughly in this order. Approximate time budgets are suggestions, not rules — weight your talk toward what's most substantive in your project. ### Introduction and Motivation (~3 min) What is the problem you're investigating, and why does it matter? Give enough context that someone unfamiliar with your application domain can follow the rest of the talk. You don't need to review all related work in detail, but situate your question in its broader context. End your introduction with a clear statement of your research question. The audience should leave the introduction knowing what you set out to find out. ### Data (~2–3 min) Describe the dataset(s) you used: where the data comes from, what it contains, how much there is, and any relevant preprocessing steps. If you had to make meaningful choices about how to clean or structure the data, say so. Be specific. "A dataset of medical records" is less useful than "a publicly available dataset of 5,000 radiology reports with radiologist-annotated labels." If there were interesting challenges in obtaining or working with your data, mention them briefly — data work is part of the project. ### Hypothesis (~1–2 min) State your testable hypothesis explicitly. Your audience should be able to hear it and know what evidence would support or refute it. If your hypothesis evolved from what you wrote in M3, you can briefly note that. ### Methodology (~5–6 min) Describe how you conducted your experiments. Your methodology section should answer: - What models, algorithms, or techniques did you use? - How did you apply data-efficient ML methods (this is the course topic — make this connection explicit)? - What were your experimental conditions or comparisons? - How did you evaluate your results? You don't need to describe every implementation detail, but you should give enough that a classmate could understand what you did and why. ### Results (~4–5 min) Present your results with visualizations where possible. Tables, graphs, or figures make results easier to follow and more compelling to an audience. Present results objectively — describe what you observe without editorializing yet. "Model A outperformed Model B by 3 points on metric X" is a result; "surprisingly, Model A worked much better" is analysis (save that for the next section). If your results are preliminary, incomplete, or negative, present them honestly. Negative results are results. Unexpected findings are interesting. What matters is that you've run meaningful experiments and can show what happened. ### Discussion and Analysis (~3–4 min) What do your results mean? Connect your findings back to your hypothesis — does the evidence support it, refute it, or is it inconclusive? If the picture is complicated, say so and explain why. This is also the place to discuss limitations, surprising findings, and what you'd do differently or next. Some of the most valuable scientific content lives here. --- ## Presenting Work in Progress Presentations are scheduled before the final writeup is due, and your project doesn't need to be complete by then. But you should have enough done to speak substantively to every section above. "We haven't gotten to results yet" is not acceptable at this stage — you should have at least some experimental results to show and discuss, even if they're preliminary. Use feedback from your presentation to improve your final writeup. Classmates and I may raise questions or point out gaps you hadn't noticed — take notes and address them. --- ## Slide Design A few practical notes: - **Keep slides sparse.** Dense text slides are hard to read and encourage you to read off the screen. Aim for one idea per slide, with visuals doing as much work as possible. - **Make figures readable.** Axes should be labeled, fonts should be large enough to see from the back of a room, and color choices should be legible. - **Don't skip the methodology.** It's tempting to rush through "how we did it" to get to results, but a clear methodology is what makes results interpretable. - **Include a slide with your hypothesis.** Put it on the screen so the audience has a reference point when you get to results. Submit your slides (as a PDF) to Blackboard by the morning of your presentation day. --- ## Questions The 5-minute Q&A is a normal part of a research talk. A few things to keep in mind: - It's okay to say "I don't know." "That's a good question — I'm not sure, but here's how I might investigate it" is a perfectly valid answer. - If a question is about something you haven't tried yet, you can acknowledge it as a direction for future work. - If a question is unclear, ask for clarification before answering. --- ## Logistics - Talks are 20 minutes + 5 minutes Q&A. You will be cut off at 25 minutes, so practice to time. - The presentation schedule will be announced in advance — let me know if you have conflicts with any of the April 28–30 dates. - Submit slides as a PDF on Blackboard by the morning of your presentation day. - All group members are expected to be present and to contribute to the talk.