CS 294-105: Class Presentations


The bulk of the course is oriented around class discussions of dataset presentations made by each student. The goal and expectation is that everyone will over the semester both present an empirical analysis and thoughtfully engage in commenting on the presentations of others. Each presentation will provide opportunities to explore different facets of the analysis process, as well as to revisit themes we have previously seen in other contexts.

It is important and valuable to ensure that discussions are constructive and thought-provoking, avoiding overly critical remarks or tone.

What Measurements to Use?

Presentations can be based on measurements and analyses rooted in any of the following:
  1. A research project you are currently working on.
  2. A research project you completed in the past.
  3. A published paper by someone else. If possible, this means you have access to their measurements (perhaps by asking them, though it's rare that that works if they haven't already released the data). It's still doable without their measurements, though, if you focus on illuminating what the paper does and doesn't tell us about the data, along with the corresponding analysis process - making sure that there's enough "fodder" to talk about.
  4. Publicly available measurements. For these, you'll need to formulate apt analysis targets (see below), since those don't generally accompany public datasets.

For all of these, the data should be of a significant volume - at least thousands of measurements.

What Are the Overall Steps?

The process works as follows:
  1. By Friday Sep 26, send me via email (with "Presentation" in the Subject) the following regarding your presentation:

    1. The measurements you have in mind, including how they relate to the different types sketched in the previous section. Explain their nature and volume.

    2. The analysis targets for the measurements: what sort of questions will you try to explore/address using the data? For example, for Homework #1 the analysis targets were whether the arrivals of individual users are well-modeled as a Poisson process, and a similar question for the keystrokes of individual users. Note that it might work fine to have a single target, if it's adequately "meaty" in terms of the amount of investigation it requires.

    3. When you would ideally like to present (early, middle, or late in the semester). Keep in mind that later presentations will be expected to deliver a higher level of analysis since they benefit from the knowledge developed from the earlier presentations.

    4. Any strong constraints you have on presenting (times you'd really like/need to avoid).

    5. Whether you can share your data with the rest of the class. (This is strictly optional.)

  2. One week before your presentation, post draft slides to Piazza, along with a sketch of your problem domain, the nature of the data, and analysis targets, if these aren't already clear from your slides.

  3. If you can share your data (or a subset) with the class, also post it on Piazza.

  4. Present the data in class, aiming for about 40 minutes total including discussion. (See the next item for what goes into a presentation.) We have enough students that we'll aim for one presentation by each student over the course of the semester, which will make for two presentations per class meeting.

  5. Engage in subsequent Piazza discussion regarding investigatory questions that came up during your presentation. To the extent that you can, endeavor to help "close the loop" on ideas that come up during the class's exploration of your data.

What Does a Presentation Look Like?

In your presentation, you should develop the following elements:

  1. The raw data: its nature, volume, duration, how it was gathered, considerations about the measurement process.

  2. Data quality: steps you took to assess any potential problems with the data and what these revealed, including biases or limitations.

  3. Exploration: how you explored the characteristics of the data and what you found. Note that this stage can feed back into considerations about data quality.

  4. Analysis targets: for each of your analysis targets, discuss how you pursued it.
    Argue your conclusions, meaning consider how you would endeavor to present the conclusions succinctly but persuasively in a paper.

What If I Still Have Questions?

Ask away! Post to Piazza, or if you prefer, send me email or talk after class or at office hours.