At work, we recently started biweekly State of Machine Learning (SoML) sessions. The general idea is to make every ML team member know what problems we are working on and what our solution approaches are. During the session we present short summaries of current work one by one and gather feedback from the rest.
We have been trying to derive some sort of consistent value from these team wide meetings earlier too. Past attempts included a process meeting (talking about improvements and automation in our workflow), a problems meeting (listing down problems) and many more. Most of these focused on specific aspects that went up and down in priority which, perhaps, made them go out of fashion soon.
With SoML, we are looking at something similar to what is called a Design Critique in the design world. This is a little different from review as we are less focused on being a step in the formal delivery process. We, of course, don't go very formal with the critique process in SoML but the general decomposition of the session semantics looks similar.
After merging a few ideas, here is the list of things that should happen via SoML1:
- General updates for the whole team
- Discussion of better approaches for current problems
- Cross project engagement at a higher level of abstraction
- Learnings
All have durable value. If communicated well, theses sessions should work out for longer than our earlier attempts.
Footnotes:
We might rename it to ML Critique in time.