Credit: Ryan Garcia

The Self-Organized Movement to Create an Inclusive Computational Neuroscience School

In less than three months, Neuromatch Academy developed an interactive online course catering to more than 1,750 students in more than 70 countries.

This past May, Panagiota (Yota) Theodoni, a postdoctoral fellow at New York University, engaged in an epic email sprint, sending more than 100 emails to 100 different countries. As an organizer for Neuromatch Academy (NMA), a fully online computational neuroscience summer school, Theodoni was determined to recruit a diverse group of participants. “We made sure to reach out to every country on this planet with at least one message,” she says.

Theodoni’s team wrote emails and Facebook messages and filled out website contact forms in many different languages. They ultimately contacted 195 countries and recruited participants from more than 70, ensuring that NMA would reach more of the globe than any neuroscience school before it.

The course, which catered to more than 1,750 students over three weeks in July, arose out of necessity, filling a void left by the pandemic-fueled cancellations of schools, conferences and workshops. But Konrad Kording, a neuroscientist at the University of Pennsylvania, also saw the virtual format as an opportunity to address some of the problems with traditional summer schools, such as lack of diversity.

For decades, specialized summer schools have provided important learning and networking experiences that catapult trainees into new technical territory and connect them with researchers in their field. But these experiences are unattainable for most of the world — in-person summer schools are costly, require extensive travel and typically accept a small fraction of the students that apply. “As someone from Pakistan, attending a physical summer school is almost impossible because of the costs as well as the visa issues,” says NMA participant Salman Maqbool. Participants at in-person courses in the U.S. and Europe have also been historically very male and very, very white.

Many leaders in the field have pushed for remote conferences and schools in recent years, motivated in large part by a desire to reduce travel and temper climate change. The extreme limitations brought about by the pandemic finally rendered these plans not just desirable but necessary.

Kording already had experience quickly pulling together online events. Early in the pandemic, together with Dan Goodman, Titipat Achakulvisut and Brad Wyble, he developed an online ‘unconference,’ which featured both lectures and a virtual networking component designed to mimic the in-person interactions that make conferences so valuable. (For more, see “Designing a Virtual Neuroscience Conference.”) Soon after, they decided to spin that success into a full-fledged summer school offering live lectures with top computational neuroscientists, guided coding exercises to teach mathematical approaches to neural modeling and analysis, and community support from mentors and teaching assistants (TAs).

The result was a summer school with well-designed content, a diverse student body, including participants from U.S.-sanctioned Iran, and a determined group of organizers who managed to pull off the most inclusive computational neuroscience school yet. NMA now has its eye on a future with even broader representation across countries, languages and skill levels. This year has been incredibly difficult for many, but NMA has provided an important precedent for how to collaborate across, and even dismantle, all sorts of barriers.

The structure of an online school

The traditional model of a neuroscience summer school involves flying somewhere and spending two or three intensive weeks working alongside fellow students. Though deeply enriching, these courses are also extremely competitive and can cost thousands of dollars, putting them out of reach for many. With these barriers in mind, NMA built a course that was fully accessible to anyone with an internet connection.

The heart of NMA is the carefully created and edited content, including prerecorded video lectures and interactive coding tutorials, which are now freely available online. Lectures and tutorials were written by a team of volunteer experts following a style guide and edited by even more volunteers for accuracy and clarity. The process was grueling but resulted in high-quality, vetted content. Sean Escola, treasurer and fundraising lead for NMA and a neuroscientist at Columbia University, spent 20 hours on the first version of his 30-minute lecture and another 20 hours revising it based on the feedback he received. “It’s definitely the most polished pedagogical content I’ve produced,” he says.

A Spanish-speaking Neuromatch pod meets for a virtual discussion. Alex Gonzalez

Participants — mostly graduate students, along with some undergraduates and postdoctoral fellows — were split into more than 180 different ‘pods,’ loosely organized by experience level and led by a paid TA. (Funding from the Simons Foundation and others helped support the course.) A typical day included six or more hours watching lectures, working through the related Google Colab notebook tutorials and attending pod discussions led by a TA. Smaller groups within pods also worked on projects, which were expected to take another two or so hours a day.

Students could also join NMA on the observer track, interacting with the materials and discussion boards in their own time but not joining a pod. After the course started, students who couldn’t participate in the live event organized their own pods to discuss the material.

Launching and running the school required an army of volunteers, working together via Slack on many independent teams. Theodoni was one of roughly 200 volunteer organizers who contributed a collective 16,000 hours and helped the course get off the ground in less than three months.

Many volunteers were brought in by the NMA board to do a small task, with the promise that “it won’t be much work.” The Princeton postdoctoral fellow Tara van Viegen, for example, was asked to caption the videos. She and the NMA video production crew took this call to action and ran with it, generating captions in three different languages — English, Spanish and Mandarin. Many of the volunteers have similar stories. Kording pitched the idea of NMA to them, and several months later they found themselves building an entirely new computational neuroscience school.

Independent groups of volunteers handled content creation, technical issues, fundraising, community building, public relations and dealing with political sanctions. “There are things that were happening that I did not find out about until five or even 10 days after they started — at which point it was completely done,” says Megan Peters, a neuroscientist at the University of California, Irvine and chair of NMA. It’s clear NMA succeeded not in spite of its scrappy, grassroots nature, but because the apparent chaos beckoned organizers to keep the engine running.

In the end, somehow all of the pieces fell into place, sometimes at the very last minute. “Have you seen Wallace and Gromit?” Peters asks. “There’s this scene where Gromit is riding a tiny little train and is putting the track down like immediately in front of the train — that is exactly how NMA felt.”

Maximizing inclusivity

NMA’s organizers were determined from the outset to make the course as inclusive as possible across gender, socioeconomic status, language and geography. NMA accepted 80 to 90 percent of applicants, and many more people participated as observers. Applicants were only excluded if the NMA team felt they could not keep up with the program.

First, they tackled gender representation. Although computational neuroscience conferences like Cosyne typically have about 17 to 18 percent participation from women, roughly half of NMA students and a third of all NMA TAs, lecturers and panelists were women. “We did this by emphasizing recruitment of women for all positions,” Wyble, a psychologist at Pennsylvania State University and an NMA co-chair, told the audience during NMA’s Inclusive Neuroscience session. (For more on bias at Cosyne, see “Addressing Gender and Institutional Bias at Cosyne.”)

With racial and ethnic diversity, the breakdown is less clear. The NMA board felt that it was too complicated and potentially illegal to directly ask an international group of participants about their race and ethnicity in a way that would be inclusive of all regional considerations. Instead, they asked participants how they would describe themselves more broadly in a post-course survey; they are sorting through that data now.

To make the course widely accessible, NMA organizers also made sure to keep costs low. The base fee for students was $100, but they adjusted the fee for the median income for a given country and offered waivers for students as well. About 15 percent of students had their fees waived.

Many international scientific conferences are conducted in English, the default language of science. To encourage a more diverse pool of participants, the NMA team opted for a broad panoply of languages. They recruited multilingual pod leaders and video translators, running pods in 13 different languages.

“English is informally the second language in Pakistan, so I didn’t have much trouble, but I really appreciate that Urdu was provided as one of the language options for the pods,” Maqbool said. “I think that may encourage a lot more people from Pakistan to apply.”

Community building was an intentional component of NMA design. “We actually made a decision early on to use language that was about community,” Escola says. “I think that’s really important as an organization because it reinforces and highlights our values and mission.” The message from the organizers was clear: “We are here with you, we are all building this together, and you are a part of this community,” Peters says. “There was no other way to talk about what NMA was and what we were trying to accomplish.”

NMA offered students informal opportunities to connect, such as virtual karaoke. Neuromatch

Despite operating in their own time zones and in language-specific pods, students had opportunities to connect through the NMA discussion boards, Mozilla Hubs and social activities, such as yoga and karaoke. Though online learning can often be isolating, NMA’s organizers, TAs and students expressed a deep sense of community. It’s impossible to fully re-create the informal interactions of an in-person course, which often happen late at night or over a beer, but these activities at least approximated the social atmosphere.

The NMA organizers’ commitment to inclusivity and community was put to the test one week before opening when they realized that U.S. sanctions against Iran prohibited training or payment to any student or TA based in Iran. In a string of tweets and public announcements that would be any PR person’s nightmare, they had to tell their Iranian participants that they could no longer attend NMA.

Amid calls on social media to shut down the academy altogether, the NMA team decided to organize against the sanctions rather than close the school. “They tried to make a change, they dealt with all the community anger, and they reacted in the best way,” says Elnaz Alikarami, an Iranian dentist turned neuroscientist who played a large role in coordinating NMA’s effort to overcome the sanctions. Through their legal team, NMA filed for an emergency license with the U.S. Office of Foreign Assets Control. This type of license typically takes months or years to process. But NMA’s license was approved in five days, just three hours before NMA was scheduled to begin. In those short hours, the NMA team scrambled to get about 50 Iranian students and eight TAs back on board. “It was kind of miraculous,” Alikarami says.

Those efforts will likely have an impact beyond the course itself. “I wasn’t sure about applying for a Ph.D. due to the challenges of immigrating, but after NMA I realized I’m really curious about neuroscience and should stop restricting my life to Iran,” says Arefe Farahmandi, a TA from Iran. “Communicating with different nationalities gave me tremendous joy that I’ll never forget.”

Technical challenges

Launching a large-scale, virtual international school comes with serious technical challenges. NMA co-chair Wyble likened it to “a firehose of problems that you never anticipated.”

NMA ran 24 hours a day for three weeks. This is a screenshot of one of the team’s many organizing spreadsheets. Neuromatch

Live discussion sessions were split into three primary zones: Asia, Americas and Europe. This meant that something was happening around the clock. Peters hosted Q&A panels at 11 p.m., and NMA Slack channels were constantly buzzing. “The last thing you do before going to bed is check your Slack notifications, and the first thing you do in the morning when you wake up is check your Slack notifications,” Wyble says.

Organizers had to deal with country-specific technical issues. For instance, many tools, including Zoom, the payment portal, and Google products, do not work in China. Patrick Mineault, a computational neuroscientist and NMA’s chief technical officer, led a team that adapted all the NMA platforms for China-friendly sites such as Bilibili or WeChat, so that 200 or so students from China could participate. On top of this, many U.S.-based companies, such as GitHub, are inaccessible to Iranian residents.

Simply sharing content presented its own challenges. Michael Waskom, a postdoc at NYU, a Simons fellow, and creator of the popular visualization platform Seaborn, helped lead tutorial editing and publishing. “In a sense, the main achievement there was that nothing broke,” he says. Most of his effort went into predicting the sorts of technical challenges that might arise due to distance and scale.

“In an in-person summer school, the person who wrote the tutorial is probably in the same room as the students,” Waskom says. “If there’s some bug that is preventing the code from running, having everyone in one room means that it can probably be easily resolved and then fixed for everyone.” With thousands of users distributed across the globe, implementing fixes is much more difficult.

In one module, for example, about 1,000 students needed to simultaneously download a dataset from the Open Science Framework (OSF). Waskom was out walking his dog when it occurred to him that this might overload the OSF server. He immediately sent a Slack message to the technical crew about it. By the time he returned home, they had assessed the potential issues and devised a few workarounds.

Captioning very technical videos in different languages was also a major task. Videos were automatically captioned but required careful proofing. In one case, the iconic “Purkinje neuron” was translated as “kitchen neuron.” Caption leader van Viegen and her team spent hours editing captions — a 10-minute video typically took 30 minutes.

The NMA team is already thinking about improvements for next year’s course, beginning with translations for their video content, additional content tracks for different levels of difficulty, and better outreach to more countries. “At the end of the day, our most represented city was London,” Escola says. “We didn’t want to exclude all of the [University College London] kids from our program, but I think we can do better.” By using the international network they built this year, they’re hoping to further even out the global distribution.

As for advice to folks in other fields who are hoping to run a similarly successful conference, Peters jokingly suggested making the same false promise that lured her in: “It won’t be much work.” After a hearty laugh with the other NMA organizers on our call, she added, “It’s a lot of work, but it’s worth it.”


Recent Articles