At the Vector AI Job & Data Fair, the auditorium is packed with hundreds of the top AI students from around the world. They’ve come to attend the Deep Learning & Reinforcement Learning Summer School in Toronto and this is the marquee event. With company booths set up for tech firms, industry labs and research hospitals, conversations tonight could be the start of solving major questions in healthcare, finance and beyond.
But the greatest draw is the three AI pioneers who are about to take the stage – Geoffrey Hinton, Yoshua Bengio and Richard Sutton. The CIFAR fellows are considered the godfathers of AI and need no introduction to this crowd.
These three scientists inspired many of the graduate and post-graduate students in the room to pursue careers in AI research in Canada or to come from abroad to attend the summer school.
Hinton began his talk with the history of neural networks – an area of AI inspired by the human brain. For a long time, neural networks were brushed aside even by other AI researchers.
“The AI people wanted to solve reasoning while the neural net people wanted to solve biology,” said Hinton. “It was a battle that went on for 60 years.”
Hinton and his colleagues persevered and as computing power improved, neural nets began to work and ushered in the deep learning revolution.
It’s happening now,” says Sutton. “And it’s happening here in Canada. Isn’t that amazing?
In 2017, the Government of Canada announced a $125-million Pan-Canadian AI Strategy led by CIFAR. The strategy supported the creation of three new AI institutes – Amii in Edmonton, the Vector Institute in Toronto and MILA in Montreal – that build on the hubs of expertise Hinton, Bengio and Sutton established in machine learning over decades.
As global interest in AI intensifies, the summer school remains a unique and important place for the next generation of AI researchers.
It was perhaps best exemplified as Hinton, Bengio and Sutton answered questions from students. Hinton shared his new thoughts on thought, Bengio weighed the benefits of academia and industry and Sutton shared his vision of the future of AI. Their advice was candid, funny and felt as intimate as the smaller summer schools of years past.
The first NCAP CIFAR summer school
In 2005, CIFAR’s Learning in Machines & Brains program (then known as Neural Computation & Adaptive Perception) hosted its first summer school. It brought together CIFAR fellows and their trainees from Canada, Israel, Finland, Scotland, and the U.S.
Hinton wanted to model the summer school on the Connectionist Summer Schools he had organized with CIFAR Advisor Terry Sejnowski, which were influential in the development of neural network research. At the time, deep learning approaches had not yet made it into mainstream university curricula and were not well represented at AI conferences. The school’s goal was two-fold: advance research and foster the next generation of AI scientists.
With the 2018 Deep Learning & Reinforcement Learning Summer School, it’s clear both goals have been surpassed wildly. The summer school has grown from a small meeting of about 30 researchers to a class of more than 250 students from 20 countries. This cohort was selected from a pool of 1200 applicants and represents some of the most highly sought after researchers in the world.
At its core is the technical curriculum and the exposure to new approaches. The summer school boasts an all-star lineup of professors from the AI institutes, industry labs and universities like MIT, Carnegie Mellon, Cornell and Princeton.
“I think it was really important for students to make that personal connection to some of the mentors they had seen in videos before or maybe they had read their textbook. But now they have a chance to meet them face-to-face,” said Taylor, who is a faculty member at the Vector Institute and the University of Guelph.
Meeting fellow students is just as important, he says, noting that he met some of the people who he collaborates with now at the summer school.
Alumni of the summer school can be found leading the top universities and tech firms. Past attendees include Roland Memsivic, who founded TwentyBN, Ruslan Salakhutdinov, who is now Director of AI Research at Apple, and Ilya Sutskever, who co-founded OpenAI.
Many of this year’s organizing committee are alumni success stories themselves. CIFAR Program Co-Director Hugo Larochelle attended the first summer school as Bengio’s student at Université de Montréal. Now he is Research Scientist at Google Brain and Adjunct Professor at Université de Sherbrooke. He and Taylor share similarities in their academic careers: moves to the U.S. and return to Canada as passionate champions of its ecosystem.
The next generation
The next generation of AI researchers are building their own stories. Some of them are students who decide to stay in Canada, others are coming from abroad to study at Canadian institutions, and some will work around the world.
The talks by students offered a glimpse of the future. Farzaneh Mahdisoltani, a PhD student at the Vector Institute and AI researcher at TwentyBN, presented her work on deep learning and video understanding. She showcased her research on a new dataset which helps computers differentiate between videos of very similar actions like putting an object down or pretending to put an object down.
I think people will really remember that Canada was at the frontier of it.
Even in the talks of three students a wide range of research areas and experience from at least eight of the top institutions in Armenia, Canada, England, Iran and the U.S. were represented.
Canada and the summer school have long been a beacon for global AI talent. The hope now is that with greater opportunities and a thriving ecosystem, even more students will see Canada as a place to advance their education and pursue careers.
For Thor Jonsson, Canada’s reputation led him to move from Iceland to study deep learning. He is a Master’s student with Taylor at the University of Guelph and plans to do a PhD in Canada.
Jonsson said he has been inspired by the Canadian AI ecosystem and believes the future is even brighter.
“I think people will look back in 100 years’ time and think about this as a historical event in scientific history,” he says.