For Canada CIFAR AI Chair Chris Maddison, a return to the birthplace of deep learning means more collaboration, creativity, and curiosity.
Photo courtesy of Chris Maddison
“Coming to Canada was a no-brainer.”
Arizona-born Chris Maddison’s job is to think deeply, but he admits that his decision to return to Toronto, where he completed his undergraduate and masters degrees, was easy. He cites personal reasons being one draw—the many friendships he developed and the vibrancy of the city are certainly attractive—but the stronger pull was professional. For him, Canada is the centre of the AI universe.
“I’m excited about the people in Canada,” he says. “It feels like every week some interesting idea comes out of the Toronto machine learning group, and I want to be a part of that. I want to interact with the faculty and the students. I think it's a really exciting place to do machine learning today.”
“It feels like every week some interesting idea comes out of the Toronto machine learning group."
A rising star in the field, Maddison studied for his PhD in 2019 at Oxford University, where he also worked with DeepMind as a founding contributor to the AlphaGo project. Currently a member of the Institute for Advanced Study (IAS) in Princeton, New Jersey, he will start a research lab in Toronto in 2020, affiliated with the University of Toronto and the Vector Institute as a Canada CIFAR AI Chair.
“The funding system in Canada has traditionally supported curiosity-based research,” says Maddison. “This is a huge advantage when it comes to fields like artificial intelligence and machine learning, because these fields have not settled paradigmatically. There are still open, foundational questions not just about how to solve well-specified problems, but open questions about what's relevant.”
This support for curiosity-driven research, combined with the collegial and collaborative atmosphere of the city, is what Maddison believes makes Toronto the best place to continue his research on the fundamental underpinnings of machine learning. According to Maddison, the combination makes Toronto “of a kind” with the Institute for Advanced Study and Oxford, the other centres of transformative research where he has worked.
Teaching old algorithms new tricks
“Machine learning is this hot new field, and we're tackling a lot of interesting problems that are exciting to talk about,” says Maddison. “But if you dig into the machinery of the algorithms that support machine learning and AI today, it turns out that the algorithms that we use are algorithms for some very basic, old problems that people have been studying for a long time, like optimization and integration.”
Maddison hopes to uncover new insights about these centuries-old problems through the study of machine learning. “We can learn new things about these old problems because we have new demands and perspectives that weren't considered in the past,” he says.
"We can learn new things about these old problems because we have new demands and perspectives that weren't considered in the past.”
For example, one of Maddison’s research projects is to understand the limits of optimization algorithms.
“In the classical study of optimization there are different categories of algorithms,” he explains, “and each category has an increasing computational expense. There are zero-order, first-order, and second-order optimization algorithms. Machine learning has, with other fields, helped drive a renewed interest in zero- and first-order optimization algorithms, because of the scale and type of problems we are solving.”
The most popular optimizer in the deep learning field right now, used for many computer vision and natural language processing tasks, is called the Adam optimizer. It was developed by fellow Canada CIFAR AI Chair Jimmy Ba, and most people consider it a first order optimizer. Adam is a very fast and useful algorithm, but Maddison wants to know if it’s possible to do better.
“Is Adam the end of the story?” he says, “Perhaps, but I’m not completely convinced.”
Exploration and possibility
Maddison is passionate about the importance of intellectual freedom for exploring ideas. Having the space to think creatively and talk with colleagues, without the pressure of immediate applications, leads to the flashes of insight that exhilarate him and transform the world. Support through the Canada CIFAR AI Chairs program will enable him to continue pushing the boundaries of what is possible.
“It can be scary, but many of our most exciting ideas have come from that space of exploration and possibility.”
“I think it's important to have that little corner of your research agenda that is essentially a dream, in which you're not exactly sure where you're going or if you’ll succeed,” he says. “It can be scary, but many of our most exciting ideas have come from that space of exploration and possibility.”
Time is another essential ingredient for discovery. “Usually one gets stuck when doing research, and that’s when I put the more fanciful dreams aside, to let them hibernate. But there have been a couple of moments in my life when, for reasons out of my control, I could suddenly see how those fanciful dreams might be turned into new algorithms,” says Maddison. “Those are the moments that drive my passion for the field. That's why I love doing this.”
The Canada CIFAR AI Chairs Program is the cornerstone program of the CIFAR Pan-Canadian AI Strategy. A total of $86.5 million over five years has been earmarked for this program to attract and retain world-leading AI researchers in Canada. The Canada CIFAR AI Chairs that have been announced to-date are conducting research in a range of fields, from machine learning for health, autonomous vehicles, artificial neural networks, climate change and more.