Wiki Contents

Reading Group

Purpose

The KDD Lab reading group discusses research papers in its foundational research areas of machine learning representation for: natural language; computer vision; reinforcement learning; cybersecurity and privacy-preserving AI; and other applications such as data mining.

Dates, Time, and Venue

The reading group is currently held online in the KDD reading group Zoom channel every Monday at 1500 CST (the time slot for spring 2024) and the meeting time is determined at the end of each previous semester using a poll. The reading group meeting recordings can be found here.

Upcoming Papers

Link for Mobile Devices

Responsibilities

  1. Select the paper you are going to discuss **2 weeks in advance**
  2. Keep your column in the [Papers tab of the reading group spreadsheet](https://docs.google.com/spreadsheets/d/1ZA7fpjw6BQBZuqw3ntlsKBv8mb1uNtifTTeZLPGIz8s/edit#gid=327559737), which is a queue of papers you are planning to lead a discussion on (first paper) or considering as a nomination for the reading group (rest of your list), filled with *at least 5 papers*.
  3. Announce the paper you are going to lead a discussion on in the spreadsheet and let Yihong know. She will announce the next reading group meeting using a Google Calendar invite and in Slack (you can set these to notify you by e-mail if you wish).
  4. Read the next week's paper.
  5. Come to the reading group prepared to discuss the strengths, weaknesses, and context, of the paper (see Expectations below).
  6. Continue your literature review to choose your next paper every week between advising meetings. *Discuss papers you have found and are considering with Bill.*

Expectations

  1. Attend every reading group you can.
  2. Come prepared to participate in discussions. Think about strengths, weaknesses, and context. Context includes use cases; seminal, methodological, or applied nature of the paper; the caliber of the venue (conference or journal).
  3. Participate. Speak up about your critical review of the paper. Ask questions.

Paper discussion criteria

  1. Interest: #general, #machinelearning, KDD core divisions (#vision, #nlp, #reinforcement-learning)
  2. Soundness of theory
  3. Use cases and examples, corpora, and available data (open or otherwise)
  4. Implementations (open source or otherwise)
  5. Reproducibility of results
  6. Open questions or unclear points
  7. Quality of the venue, especially relevant venues to publish in or ready
  8. Quality of the paper
  9. Key strengths of the paper, especially to emulate where possible
  10. Key weaknesses of the paper, especially to avoid where salient
  11. Ideas that arise from or are inspired by the paper, including take-home concepts
  12. Follow-up, especially outreach to the author(s) and understanding the review history of the paper

Next Paper

Date: week 0 or week 1 of spring 2024
Location: Zoom
Time: TBD
Paper: Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback by ...
Source:
Scheduled Discussion Leader: William Hsu (postponed from fall 2023)

Last updated by bhsu on Nov 14, 2024