The University of Waterloo, known for its intensive STEM focus, has graduated big tech names including PHP coding script creator Rasmus Lerdorf and BlackBerry founder Mike Lazaridis. The school is located in the Kitchener-Waterloo area, two hours west of Canada’s largest city, Toronto. It may be out in the countryside, but it’s become a hot spot on the Pan Canadian AI Strategy map, and Toronto’s “AI twin” devoted to applied research.
The university launched the Waterloo Artificial Intelligence Institute this April, pooling 8 affiliated research centers, 23 labs, and some 100 faculty members into a new Canadian AI supercluster with a focus on operational AI research.
UWaterloo Ties with AI: Speech Processing & NLP Pioneers
UWaterloo’s ties with AI date back to 1989, when, supported by funding from Nortel Canada, Bell Canada, NSERC, the Ontario Government, Canadian Department of National Defense, and the US NSF, Professor Li Deng led a group of 20 students and several postdoc fellows in work on AI.
Along with Professor Sherman Shen, Deng published papers using statistical methods such as HMMs to tackle speech recognition, NLP, and signal processing problems. UWaterloo researchers also dabbled in neural networks with Professor Mohamed Elmasry, but that work was stymied by hardware and data limitations of the time.
Deng left UWaterloo for Microsoft Research in Seattle in 1999. In 2012 he collaborated with deep-learning pioneer Geoffrey Hinton on the important Deep Neural Networks for Acoustic Modeling in Speech Recognition.
At around the same time, UWaterloo computer science students Sam Pasupalak and Kaheer Suleman met for the first time in an AI class. Sharing a passion for machine understanding, the two founded the startup Maluuba in 2011, securing seed funding from Samsung Ventures. In just six months, Maluuba had built an Android voice assistant application that rivaled Siri and Google Now.
Riding the success of deep learning and supported by esteemed advisor Yoshua Bengio, Maluuba was an up-and-coming NLP star, amassing CDN$11 million in funding and setting up a research lab in Montreal. In June 2016 the team created a machine comprehension system, EpiReader. Trained with CNN/Daily Mail news and children’s storybook datasets, EpiReader was able to fill in blanks using contextual information, outperforming industry leaders DeepMind and Facebook with 74 percent accuracy in machine comprehension tests.
Maluuba’s strong research capabilities and bold innovations did not go unnoticed, and by early 2017 the company was acquired by Microsoft. Maluuba went on to release the NewsAQ and Frames datasets, which could be used to train deep-learning algorithms to develop reasoning skills and conversations. Microsoft meanwhile continued making NLP startup acquisitions to feed its conversational AI ecosystem.
Creating a New Artificial Intelligence Institute
UWaterloo Cheriton School of Computer Science Professor Peter Van Beek joined the university in the early 2000s, soon after Deng’s departure. Today he co-directs the new AI Institute with Professor Fakhri Karray from the Department of Electrical and Computer Engineering.
“[The AI institute] happened here organically,” UWaterloo President and Vice-chancellor Feridun Hamdullahpur told Synced at the institute’s April 6th opening ceremony. “It was an internal decision to set up the institute and we are very much project based. The Deans of CS and engineering gave the institute money for five years, and we are also raising money from the industry.”
Institute Co-director Fakhri Karray leads the Centre for Pattern Analysis and Machine Intelligence (CPAMI), which recently developed an AI-augmented camera system that can identify texting, talking and distracted drivers. The center also developed a similar driver behavior recognition system that tracks eye movements to warn sleepy drivers.
One of the Principal Investigators at UWaterloo today is Professor Alexander Wong, who did his undergraduate studies at the school and has been teaching there since obtaining his PhD in 2010. He leads the 40-member Vision and Image Processing (VIP) research group, which created an AI-powered imager used to monitor and predict cardiovascular disease, an AI-powered deep tissue scanner for skin cancer and other diseases, and an AI that builds effective and efficient deep neural networks. The projects have been commercialized by spinoff companies Elucid Labs and DarwinAI.
CHIL lab, led by Professor Jesse Hoë, has about 15 members conducting research on how emotions help with human decision making. The group is building applications such as virtual assistants for smartphones, smart home solutions to help elderly affected with dementia and other cognitive disabilities, and intelligent agents for online collaborative groups and social networks.
Partnering with Industry for Application-Driven Research
The Waterloo AI Institute’s aim is to develop “disruptive technology based on foundational AI that addresses operational problems and enables new products and services.” The do so, it will need industry partners.
The Institute offers paid partnership options. In the Prime package for example (CDN$250,000 annually for at least four years), partners get a rotating position as an external industry officer on Waterloo.ai’s advisory board, access to Principal Investigators (PI) and Grad Students to address defined research problems, and so forth. Current sponsors include General Motors Canada, Huawei Technologies, RBC Borealis AI, Shopify, Loblaw Companies Limited, etc.
The school has also introduced a “reverse co-op” short training program for company employees. Courses on AI’s applied capabilities and limits will be offered in Toronto and Waterloo and run from one day to a month.
The Institute’s founding will boost the local innovation scene. Google already has some 550 employees in Kitchener’s Breithaupt Block, just a few kilometres from the University of Waterloo. Concurrent to the founding of Waterloo AI, Google announced the brand new 365 square metre Google Waterloo Community Space, which will run STEM education workshop and events. Also in the immediate neighbourhood is Communitech, an innovation hub that supports 1,400 companies.
Professor Van Beek notes that Waterloo also has a big tech partner just a couple of hours east: “Our work runs complementary with the Vector Institute in Toronto — they do a lot of foundational work, while we want to focus on application.”
As an ambitious pair of “AI Twins,” Toronto—Waterloo can be expected to accelerate the commercialization of AI in Canada.