Looking back of one season of running the WiMLDS Paris Paper Sessions
Starting from September 2019, I had the opportunity of running the WiMLDS Paris Brown Bag Lunches. Now that this has come to an end for me, I took the time to reflect on that whole adventure.
First things first: what are the WiMLDS BBLs?
WiMLDS is a global initiative, and its initials stand for “Women in Machine Learning and Data Science”. Its mission is to support and promote women and gender minorities who are practicing, studying or are interested in the fields of machine learning and data science.
In 2017, a Paris chapter was created. You can read directly from one of the chapter’s co-founders here. They started with evening meetup events showcasing speakers mostly from Paris, and went on to build a solid community around these events. To this day, they are one of the biggest WiMLDS meetup groups worldwide with more than 3500 members.
In early 2019 they started highlighting academic papers written by female authors, and sharing them on Twitter with the hashtag #VendrediLecture, a common French hashtag meaning Friday Reading.
It was a natural next step to combine these highlightings of academic papers, with the growing community of Paris machine learning and data science enthusiasts.
And this is when I joined this adventure! The organizers of the evening meetups had enough on their plate already with these events, so they asked me to come in and undertake the organization of these paper reading sessions.
Putting the paper sessions in place
For these paper sessions, we wanted to propose a new format and organize them during the lunch hour, so that people available at different times would get a chance to participate. So, I suggested that we adopt the Brown Bag Lunch format, where people gather around the lunch meal, either brought by them (thus the brown bags) or supplied by the event.
However, the middle of the day can be a complicated time frame to organize, it can conflict with people’s work schedule. Contrary to evening meetups that are a good occasion to discover and visit the offices of cool companies in the city, we thought it would be preferable to always host the Brown Bag Lunches at the same central place. This way, that place would be consistent and would not be a complication factor in the schedules. Agorize was nice enough to volunteer their offices for these events, as well as a lunch budget.
How did that turn out?
As mentioned, Agorize provided us with a lunch budget. The simplest option seemed to be pizza. In practice, it was not practical at all to eat while discussing the papers, so we actually separated the lunch time from the paper discussions. The lunch was then purely a time of socialization between attendees. It worked so well, we even had attendees who came more for that socialization time than for the papers!
As for the papers, it seemed risky to assume that everyone would have read the paper and be ready to discuss it, so I decided to start by presenting my understanding of the papers.
The goal of these gatherings has always been to be as interactive as possible. It is not as easy as it sounds!
The theoretical plan was to present a summary of the papers, and then have a discussion of what people understood, possible improvements and applications, and so on. In practice, I found it was hard to have these discussions happen organically, and we had to ask directed questions to the attendees to try to create them. I assume people are too used to the usual meetup format, where one person presents, and everyone else listens and ask questions at the end. It was hard to get out of that pattern.
After a few occurrences of these BBLs, people were starting to be a little used to it and we started to see a few regulars come back. I then tried to create interactions with the attendees differently: what if they chose papers and presented them to the audience?
In my opinion, this initiative was a success. We had a few presenters that came forward, and a few others that I contacted directly. It was a great opportunity to showcase some of the Machine Learning enthusiasts of our community. It also created a bigger diversity of the types of papers presented in the sessions, since different people were choosing them.
Speaking of diversity, we also noticed that our attendees were really different from one another. We had a big diversity of areas of interest (NLP, Recommender Systems, Image Processing….), a lot of different seniority levels (students, interns, juniors, seniors…) and even a lot of different backgrounds! In the last online sessions, we had people coming from marketing backgrounds, statistics, economy… I think it was really awesome to peak the interest of so many different people.
Meetups in the Covid time
Like every event, the paper reading sessions have been impacted by this pandemic. It was necessary for everyone's safety to move them to an online setting.
As a consequence, interactions were harder than ever to create between the attendees. People are shy, but when you recognize your regular attendees, or notice someone making faces, you can try to ask them more precisely what they're thinking. But when you only have a screen with the black boxes of all your attendees with turned off cameras… it's a whole other story!
It was a bit intimidating as a presenter to be in front of a screen with almost no faces to see. Thankfully, people quickly started adjusting to the online setting too, and they started asking questions about the papers in the chat of our visio tool. Sometimes presenters didn't see the questions right away and other attendees answered the questions. It took a bit of an adjustment period, but I think we finally found a new balance on this aspect.
There were also a few positive aspects to the online switch. We had a few attendees from other cities in France who wouldn't have been able to attend otherwise. And since we speak English during these events, we even had a few people from other countries!
What I learned
I started running these events in September 2019. We missed a few due to the public transportation strikes in Paris this winter, so this whole season of BBLs was a total of eight events. By running these events and seeing them evolve during the year, I would like to share a few takeaways.
As an organizer
People who come to the events are mostly shy and don't dare to talk a lot about their points of view. If you want to have a relatively high level of interactions, I think you have two choices.
You can have pretty good people skills and emulate these interactions yourself, by creating a dialogue between you and the attendees.
Or if you're like me, and not that socially extroverted of a person, you can try to prepare in advance a few questions and subjects to open to the audience. I tried to to take the time to present the papers to co-workers first, and note the questions that they had. The day of the event, I inverted the roles and asked these questions to the attendees at the end of the presentations.
As a presenter
One thing that was pretty obvious to anticipate as a presenter but was quickly confirmed: preparing these sessions was a lot of work!
Finding papers to present turned out to take quite some time. The preparation of the presentation was also a good piece of work. It's one thing to read a paper that peaked your interest in the subway on your way to work, and get the gist of what these people did to solve the problem at hand. It's a whole other level when you have to understand the paper enough to present it clearly to a diverse crowd of students and data professionals… and be ready for their questions.
Moreover, I was a bit worried that as a backend developer reconverted data engineer, I was going to have a really hard time understanding these papers well enough to present them. On this point, I would like to say that most likely if you work hard on them, it’s not impossible! Sure, it took me a lot of work on some papers, but I think it’s manageable if you’re willing to put the time in. I even once ran a demo project on GANs!
In the end, even if it was a good amount of work, it was also a great opportunity to finally sit down and take the time to learn about a lot of data science concepts not typically in my area.
The end
As a conclusion, it was amazing to have had the chance to start this from scratch and see it evolve during a whole year. Personal circumstances on my side mean I will not continue running these events in the fall. I’m sad to go but I’m leaving for new adventures, and I’m sure that whoever takes the lead will continue improving them again and again. I can’t wait to see these events evolve, and I’m sure the community will also be there to see what’s next.
So long, and thanks for all the fish!
Resources
I tried to sum up the content of all these sessions in GitHub pages:
b3tty.github.io/bbl-wimlds
If you want to explore directly the directory:
https://github.com/B3tty/BBL-WiMLDS
Stay informed about new WiMLDS events by joining the meetup group!
