📚 gabe's wiki

Project Presentation

Mar 19, 20255 min read

ECS 170 Project Presentation

Presentation Format:

  1. Duration: Each group will have a total of 20 minutes for their presentation.

    • Presentation Time: 80-85% of total time. Walk the audience through your project, its objectives, methods, results, implications, and what you learned.
    • Q&A Session: 15-20% of total time. This time is set aside for questions, feedback, and discussion with the audience.

  2. Division of Time: Ensure that the presentation time is distributed approximately evenly among your group members.

  3. Content Structure: Here’s a suggested flow to structure your presentation:

    • Introduction: Briefly introduce your project and its objectives.
    • Background: Recall any concepts that the audience should remember to be able to understand your project.
    • Methodology: Discuss the AI technologies and methods you applied, the challenges you faced, and how you overcame them.
    • Results: Present the outcomes of your project, supplemented with visual aids - a slideshow, and optionally a live demonstration.
    • Discussion: Reflect on the implications of your findings, any limitations, and potential future directions or improvements.
    • Conclusion: Sum up the main points and the significance of your project.

  4. Collaboration: An effective presentation will have group members primarily presenting the segments they contributed most to. However, during the Q&A, any member could be posed a question about any part of the project.

  5. Location: Presentations will be held over Zoom (or another video conferencing solution). If you would prefer an in-person presentation, please send me an email well in advance.

  6. Time: Presentations will take place during the last week of class - keep an eye on Canvas for announcement with a link to sign up for a presentation slot.

  7. Audience: The audience for the presentations will be myself and one TA. If you’re feeling confident, feel free to invite your classmates.

Preparation Tips:

  1. Rehearse: Practice your presentation multiple times to refine your flow, ensure you’re within the time limit, and bolster your confidence.

  2. Visual Aids: Use slides or other visual aids to make your presentation engaging and illustrative. Specifically, include complete figures.

    Complete figures have: - Ticks (if applicable) - Tick labels (if applicable) - Axis labels - Titles - Legends with clear legend labels - Highlights (optional)

    The mapping between marks on the figure and concepts in the viewer’s mind should be made clear by: - The figure itself, especially axis labels - Your description of the figure

    I highly recommend that you learn about the “grammar of graphics” if you haven’t already. Try searching YouTube for videos by Tamara Munzer.

  3. Engage Your Audience: Aim to communicate what you did in an accessible way. The goal is clarity and understanding.

  4. Anticipate Questions: Think about potential questions you might be asked and prepare answers. This will make the Q&A smoother and more productive.

  5. Squint test your slides: Put your laptop on a table in a small room and stand across the room. Can you read the slides? If not, the text may be too small. Beware:

  • Networks with lots of nodes and edges
  • Pairwise correlation matrices with lots of labels
  • Other
  1. Avoid code screenshots unless they are very succint (10 lines or less). Consider alternative presentation techniques for your algorithm, like high-level pseudocode or a flowchart. In short, code screenshots are not as informative as you might think. Consider how long it takes you to read source code in a library you are unfamiliar with.

What should you NOT present?

Avoid “Fluff”

  • Definitions of AI and machine learning
  • Long definitions of “what is an API”. Assume most Upper div CS majors have learned this?
  • Too many takeaways/summaries
    • One discussion and conclusion slide is ok!
    • Discussion, Takeaways, Key Takeaways, Final Takeaways, Conclusion, Last Notes is far too much!

Avoid low-information content

  • Code screenshots of >5-10 lines. Use code screenshots very judiciously or not at all.
  • Long lists of uninterpretable column names

What should you do if you’re short on time?

If you need to condense your presentation, you can trim:

  • Deep technical details
  • Intro slides like “what is machine learning”
  • Vague conclusions

Evaluation Criteria: Your presentation will be evaluated on:

  • Content depth and clarity.
  • Structure and logical flow of the presentation.
  • Engagement and communication skills.
  • Ability to address questions and feedback constructively.
  • Collaboration and equitable contribution from all group members.

Questions

Here are some typical questions you might anticipate for your presentation:

  • Is there a theoretical upper limit on performance for this task? What is it? How is it derived?
  • Is there a theoretical lower limit on performance for this task? What is it? How is it derived?
  • Are your results sensitive to your choice of hyperparameters? How sensitive? How do you know?
  • What inputs to the system influence its behavior the most? The least? How do you know?
  • How does your system compare to weak baselines for this task? Strong baselines?

Low-priority tips

  1. Don’t mix light-mode slides and dark-mode screenshots
  2. Axis labels on figures are often too small for presentation by default. You can find this out by squint testing your slides. Configure matplotlib, seaborn, altair, etc. (whatever library you use for visualization) to increase the font size.
  3. Clarify what resources you started with and what you added on top.
  4. If you have Night Light, Redshift, or another screen color temperature app on the laptop you use for presentation, turn it off during the presentation
  5. Don’t just throw an AI-generated figure in the presentation because this is an AI class. Include informative figures.
  6. Neural network training curves are usualy informative if validation and training metrics are sampled more often than once per epoch. A lot can happen in an epoch!
  7. If you include a demo video, consider showing it at 2x speed or editing non-informative segments

Good Luck!