Last September, 200 computational neuroscientists attending the Cognitive Computational Neuroscience (CCN) conference converged on a campus lecture hall in Berlin for a scientific version of speed dating. Participants had submitted an abstract describing their work, hoping to find others with similar interests. A specially designed algorithm matched each person with six others based on their abstracts. They moved from table to table, chatting with another scientist for 15 minutes before moving on to their next match.
The event, dubbed Mind Matching, was the brainchild of Konrad Kording, a neuroscientist at the University of Pennsylvania, who had been brainstorming with his graduate student Titipat Achakulvisut and then-postdoc Daniel Acuña about how to improve conferences and get attendees to interact with people beyond their typical social circles.
The event was such a success that one of the participants, Dan Goodman, a neuroscientist at Imperial College London, suggested taking it online. Goodman and others had been searching for ways to reduce the climate impact of conferences by replacing them with virtual events. Mind Matching offered a solution to a nagging problem — how to re-create the networking aspect of in-person conferences. As conferences began to be canceled in February and March due to coronavirus concerns, the effort took on new urgency.
The team quickly put together an entirely virtual conference, neuromatch, held March 30-31, that included invited and contributed talks, panel discussions, virtual happy hours, and the Mind Match pairing that inspired it all. About 3,000 people registered for the event, with as many as 1,000 viewing talks simultaneously. All the tools that researchers used to organize the conference will be open source, so that others can follow their lead.
Kording and Goodman spoke with the SCGB about the conference. An edited version of the conversation follows.
What inspired the neuromatch conference?
DG: I had been reading a lot about ideas for reducing the climate impact of conferences, but I wasn’t really convinced by them because you could replicate the talks but not the arguably more important social function. When I was a participant in the matching at CCN I was blown away and realized that this would be how you could replicate that bit, so I approached Konrad to suggest it. We started organizing the conference six months ago. Then the coronavirus hit, and we thought maybe we should do this quicker.
KK: And make it bigger.
Why is it so important to do this now?
KK: People are lonely. In hard times, we need to be there for one another. Meeting online is a way to be there for one another.
How did neuromatch compare to a regular conference?
KK: We incorporated all the components of a real conference, including talks and discussions. Mind Matching is like the networking that happens during the coffee break. We also had social events — I had been prototyping something called neurodrinking, where we assign four random people to a room to hang out with each other.
What are the advantages of a virtual conference?
DG: It can be even better than a regular conference, where everyone talks to people they already know.
KK: Extroverted people have the advantage at in-person conferences. Lots of people don’t feel entitled to go up to famous people and introduce themselves. I hope virtual conferences broaden participation, opening up opportunities for people to meet who might normally not meet. It also improves inclusivity. Sending a student to a conference often costs $2,000 between flights, hotel and registration. At neuromatch, everyone is welcome.
You originally developed Mind Matching — the ‘speed dating’ algorithm — as an in-person event?
KK: We’ve done it in person several times, and the feedback was very positive. Many people said it was great, and some thought it was the best part of the conference. The biggest complaint was that the room was too loud. With 200 people Skyping one-on-one, that’s not a problem.
DG: It was kind of extraordinary how well the in-person event at CCN worked. I sat down with two people I had never met before, both doing an almost identical project to what I was doing.
How does Mind Matching work?
KK: Matching based on keywords works poorly, so we use abstracts, which have a lot of information. We describe the abstract as a vector in 400-dimensional space — the ‘interest vector.’ We match people with nearby vectors, using an algorithm to do optimal matching. In order to avoid meeting people they already know, participants list who they don’t want to meet.
DG: It automatically selects six people to meet from the conference. People can connect with their matches whenever they like. It doesn’t have to be during the conference.
(The code for Mind Matching is available on GitHub.)
How many people signed up for Mind Matching?
KK: About 500. The bigger the pool, the more precisely you can match. The cutoff for registration for the March 30 event was March 25, so we had a few days before the conference to fact-check the matching to make sure the software was working well.
What was the biggest challenge in organizing the conference?
DG: The short notice.
KK: I am on the organizing committee of CCN. We have an organizing committee of 12 to 14 people that meets regularly for effectively the whole year, and a program committee, an advertisement committee and other subcommittees, as well as an army of student volunteers who organize everything a year ahead of time. What we have now is me and Dan, a couple of friends, and volunteers who show up randomly. We are building the entire conference infrastructure in real time.
The software is also a challenge. We were originally thinking of using Zoom, but it’s size-limited. I got Crowdcast, which scales infinitely. All the talks were recorded so people can watch on their own time.
What worked? What would you change for next time?
KK: I outlined the components of the conference that worked well in this Medium post. Here are some key points from that post:
… We found that having a break of about 10–15 minutes every hour worked very well. … During these short breaks some participants would chat with one another (we never blocked the chat) about various scientific topics. It also gave us organizers the chance to continuously optimize the parameters of the conference. …
… People could chat in realtime about the presentations. This chat channel was often even more useful than the talk itself. People were giving links to papers for background. They were answering beginner questions. They were clarifying points of potential disagreement. …
… Our conference suffered a bit from having several different communication formats (email, slack and mattermost). It would have been better had we spent some time in advance picking a platform and setting up the necessary channels.
DG: The only thing I would add is that we were really badly Zoom bombed by trolls. Zoom bombing didn’t even exist as a word as far as I know when we planned this thing. But by the time the conference was running it was endemic, and we spent hours and hours during the conference attempting to find solutions. We finally hit on something that worked in time for the last short talk session. The invited and contributed talks were fine as they were on Crowdcast.