Pre-requisites:

1. Read the paper (that includes yourself Chris Yau).
2. Read related references if you are unfamiliar with the material.
3. This is probably going to be hard, technical stuff, don't worry if you don't get it all.
4. We could use Jamboard or you will need a pen and paper and the ability to turn your camera toward your piece of paper.

Newcomers: It is not expected that everyone will be active participants in these reading groups. For newcomers to the field in particular, these meetings may feel somewhat intimidating because of the technical level of the discussion. However, research is continually evolving and the best strategy is sometimes to dive straight in! Even if the only things you get out of a reading group are a couple of terms to look up on Google later, this is a win. If you are interested then, over time and with some effort, you will catch up!

Format:

• one person who will chair the meeting
• (max) 30 minute paper introductions (recorded - with permission)
• (at least) 30 minute random discussions (not recorded)
• we will use a dual introducer model:
  • one describes the paper, the techniques and its main findings,
  • the second focuses on clarifications/benefits/weaknesses/opportunities for further development,
  • it might be useful to coordinate with your co-introducer beforehand (good opportunity to work with someone new).
• ask questions/make comments (there are no stupid questions), these sessions are only as interactive as you make them
  • e.g. can you explain more how XXX works?
  • e.g. what is XXX?
  • e.g. I work on XXX and I think YYY could be used here
  • e.g. this approach is rubbish because of XXX
• depending on numbers, raise your hand, and wait for Chair before speaking
• determine topics and volunteers to introduce next paper
• reading groups will be biweekly

Topic of the moment: What is clustering?

Background: Clustering is an ill-defined problem. Broadly, we want to identify groups of "similar" objects but how many groups? How do we measure similarity? Likely how we define and implement this is specific to each individual problem/application but are there generalities we can identify? Also, what about unusual clustering problems like picking up rare sub-populations? What about interpretability of clusters? Linking clusters to external validation data?

Papers: Papers can be from the past or recent literature or work in progress of your own. Pure methodology? application inspired? The main constraint is that it should be methodologically-focused.