Will braindrain be the continued fate of Montreal’s Machine Learning community?

Montreal’s Artificial Intelligence (AI) and Machine Learning community arguably started out as leaders in their trade, but community experts say US braindrain is slowly sucking the lifeblood out of the community.

Prior to the 1980s, Machine Learning experts were part of a small community of academics, who took advantage of plentiful government funding that helped churn out talented minds. Still, experts say both the provincial and federal governments ultimately failed to keep these treasured possessions within Quebec.

As interest in the field of Machine Learning has soared over the past few years, heavy-hitting investors are scrambling to bankroll startups, while Silicon Valley continues to swallow up talent. Google even announced its new Machine Learning platform just weeks ago.

Machine learning is the science of getting computers to act without being explicitly programmed. According to Jeremy Barnes, the CEO of Montreal-based startup Datacratic, ‘deep learning’ is a new algorithmic paradigm for Machine Learning that allows the technology to emulate some of the capabilities of humans, like image recognition, speech recognition, natural language process and more.

Machine Learning

Jeremy Barnes, CEO, Datacratic

It comes from work on ‘neural networks,’ which Barnes says will likely have a huge impact on the economy, just like the personal computer or the Internet have.

Montreal, for its part, boasts a Machine Learning lab in nearly every university in the city, while startups like Datacratic, Smooch, Fuzzy.io, Fluent.ai, and many others are leveraging this technology as their core business component.

Meanwhile, the FounderFuel accelerator program has made it known it’s open for business if AI/Machine Learning startups come calling. In their spare time, FounderFuel’s Sylvain Carle and Thibaud Marechal have also started an independent monthly meetup devoted to the field, MTL Machine Learning

According to Barnes, Montreal’s success as a centre of AI/Machine Learning revolves around the Quebec government and a data scientist at the Universite de Montreal named Yoshua Bengio, among others.

In the 1980s, “people were really excited” about neural networks. But by the mid-1990s, the field became a “graveyard.” The only places to work on it were university labs funded by the provincial and federal governments. One was Geoffrey E. Hinton’s lab at the University of Toronto, while another was Bengio’s lab at the Universite de Montreal.

“They were all financed by the Canadian and Quebecois tax payers and they were working on the algorithms back when nobody believed in it,” Barned told MTLinTech. “People wouldn’t accept journal articles they published and they couldn’t even find students that wanted to work on it.”

But as time passed, it became clear to the broader tech community that AI and Machine Learning were, in fact, exciting prospects.

JF Gagne, an Entrepreneur-in-Residence at Real Ventures who previously founded two Machine Learning startups, said Quebec’s generous funding went back as far as the 1970s.

“They were funding grants in computing, math and a few other programs, and you could actually combine all of these grants if you were doing Machine Learning or optimization. You could kind of triple-dip the grants and end up with unmatched research capabilities,” said Gagne.

Gagne sees history within a slightly more positive lens than Barnes. He told MTLinTech that over time, Quebec attracted talented students and an oversized group of people that ended up yielding strong inventions and scientific papers within the field.

Machine Learning

Yoshua Bengio

Barnes said the main problem was that Quebec’s government fell way short when it came time to keep all these brilliants minds within the province after its funding had helped spur their expertise.

“All of those people have basically gone to the Valley. While the valley didn’t bankroll the early research, they’re now all profiting from it, with the one exception of Yoshua Bengio at the Universite de Montreal. Yoshua, rather than joining Facebook or Google, or any of the other places that would have hired him in a second, stayed in Montreal,” said Barnes.

Bengio’s resume is striking. A Full Professor at U de M’s Department of Computer Science and Operations Research, head of the Machine Learning Laboratory (MILA), program co-director of the CIFAR Neural Computation and Adaptive Perception program and Canada Research Chair in Statistical Learning Algorithms.

With the rare exceptions like Bengio, Barnes called Montreal’s community a story of unavoidable braindrain. It could have been prevented by “having a government willing to support an ecosystem around that research, rather than just doing the research and letting all the benefits flow out of the province.”

Gagne called Montreal a strong community with more locally-educated and qualified people in Machine Learning or operational research than anywhere else in North America.  He referenced the fact that the world-class Conference on Neural Information Processing Systems (NIPS) was hosted in Montreal in 2014 and 2015, and will return in 2018. Meanwhile, Bengio’s lab, at over 75 students, is “unheard of” in terms of scale for that type of lab.

Still, Gagne agreed with Barnes in that the current state of Montreal’s community is a slightly pessimistic one with talent constantly in flight.

“The unfortunate thing is it’s still hard. The valley has their eyes on that talent pool and they’re trying to attract a lot of people. The numbers surrounding braindrain are insane.”

Where does Montreal’s community go from here?


Add yours
  1. 1
    Claudia Torregrosa

    Montreal has a huge potential to become an even more attractive machine learning hub. I strongly believe we could partly prevent the braindrain – at least from the most entrepreneurial individuals – once we make it clear that it is very well possible to launch successful startups from here.
    Fluent.ai (http://www.fluent.ai/) has been focusing on personalized intent recognition but there is room for a lot more, as proven by Sensing Dynamics (http://sendyn.weebly.com/) which is developing an electronic nose platform or aerial.ai (http://www.aerial.ai/) that performs Wi-Fi based activity recognition.

    In addition to FounderFuel, other players are providing access to resources that will make it possible to have more machine learning startups in town, including TandemLaunch where the three startups mentioned above are currently incubated. The great thing is that these resources are complementary.
    For instance, FounderFuel (http://founderfuel.com/) is great for existing teams that are looking for ways to create momentum, following the advice of the Entrepreneurs-in-Residence and mentors and leveraging the $50-$100k budget provided for 13 weeks.
    TandemLaunch (http://www.tandemlaunch.com/) on the other hand is more adapted to assembling a team and startup from scratch, essentially bridging the gap between patented or patentable research from university and a commercially viable product in the consumer electronics space using a $500k budget. The program lasts 18 to 24 months and makes it possible to raise Series A funding. Founders are not expected to come with their own ideas but with a very strong drive and entrepreneurial energy. The contact is also very tight with university, as professors often become Fellows (advisors) and students join as interns, notably through the Mitacs program.

    All these goes to say that Montreal definitely has a lot to offer to a variety of talented people. I am confident that we can retain some of our talent and attract more from other cities and countries if we reinforce efforts to strengthen the local ecosystem, give more visibility to the resources available, and promote existing and upcoming startups in the field: let’s make it happen!

  2. 2

    As long as Quebec continues with its anti-commercial legislation, specifically with regards to language, and perseveres with its appalling tax regime, then nothing is going to change. The solutions to these pointless problems are obvious to most and simple to implement, but it would require a lot of dumb people to smarten up immediately.

+ Leave a Comment