It’s an exciting time for machine learning research, and computer scientists at the University of Toronto are right in the middle of it.
The Canadian Institute for Advanced Research (CIFAR) named eight of U of T’s faculty members as CIFAR Artificial Intelligence (AI) Chairs in December 2019, recognizing their transformative work in the field. CIFAR pushed Canada to the forefront of AI research when it funded the ambitious Blue Sky Ideas research initiative in computer science.
The Varsity talked to three of U of T’s new CIFAR AI Chairs, all professors at U of T, about their past projects, current interests, and hopes for computer science to come.
A career in robotics and automation
When Dr. Animesh Garg, now an assistant professor at UTM’s Department of Computer Science, first started his academic career, it was as a mechanical engineer, not a computer scientist. His interest in his current field began when, as an undergraduate student in the late 2000s, he and some friends tried to build a self-driving car on a budget of a few hundred dollars.
The project set him on a path that he’s been pursuing ever since. His research mostly deals with engineering robots to automate dangerous or dull and repetitive tasks. He’s currently working with surgeons to create machines to assist in surgery. Robots can complete simple technical tasks, allowing surgeons to focus on the more complex elements of the surgery.
For all his work in workplace automation, he doesn’t believe that its impact on the job market is as catastrophic as it is sometimes made out to be. “I don’t think of machines… in any manner, replacing jobs,” he said. “We are not going to be able to completely replace a job or a position. What usually ends up happening is we try to assist that position, by… [enabling] you to extend your skills.”
He thinks that we’re going to see huge leaps in using automation to improve safety in driving and e-commerce in the near future, and that we’re going to see large-scale implementations of autonomous robotics in surgery and personal care within the next couple of decades.
Right now, Garg is developing and teaching two new courses to give students access to new and exciting learning opportunities — an advanced robotics class, complete with a new teaching lab and industrial-standard robots, and an introductory reinforcement learning class.
He’s looking forward to collaborating with other CIFAR AI chairs across the country, especially those who are doing similar research to him but starting from different fields. He thinks that collaboration of that sort is going to be a key element of machine learning research going forward.
“It could be very incorrect to say that a single person or single institution will solve all of it,” Garg said. “It is going to be a very collaborative effort, and that is what some program like CIFAR really encourages.”
Bridging theory and application: studying algorithms
Dr. Chris Maddison is perhaps most famous for his work on AlphaGo, a program that beat the reigning human champion at the board game Go. Since then, he’s been everywhere from Oxford University to the Institute for Advanced Study in Princeton, New Jersey, conducting AI research on an eccentric blend of topics.
His work revolves around some of the most fundamental problems that underpin machine learning, such as optimization, which is necessary for program decision-making. Even though a lot of his favourite research focuses on answering the general questions behind machine learning, he’s hesitant to call his research purely theoretical.
“I guess I would draw the distinction between people who do very applied work, people who do very theoretical work, and people who are developing algorithms,” he said. He focuses on developing algorithms — something that requires a fair bit of theory, but also depends heavily on the their applications.
Maddison started his academic career right here at U of T as an undergraduate student. He started studying linguistics and spent time in neuroscience, but no matter what program he was enrolled in, he kept coming back to computer science. Starting this summer, he will be returning to U of T as an assistant professor.
“I’m excited to work with grad students. I’m excited for that kind of research environment… It’s a group that a lot of people are paying attention to,” he said. He’s excited to embark on more research — most specifically, at the intersection between machine learning and efficient searching algorithms.
Maddison hopes to use his newfound position as a CIFAR AI chair to kickstart future research collaborations.
Conscious thinking by machines
Dr. Sheila McIlraith has been a longstanding name in machine learning research. She’s worked in Silicon Valley, at Stanford University, and most recently as a professor here at U of T. A lot of her recent research touches upon the idea of deliberative machine learning — how to teach computers to accomplish complex problem-solving tasks.
Researchers have made huge strides training computers to correctly perceive and recognize sensory input with machine learning, and McIlraith is hoping to apply those same principles to tasks that require conscious thinking. She’s been involved in projects that tackle this by training a computer to play Minecraft, and by putting a machine learning agent in a simulated environment with multiple rooms, so that it develops the working memory to complete a task that requires frequent context-switching.
Deliberative problems present a unique challenge to the programs that have to break down and ‘understand’ their assigned tasks. They may not be rewarded for the intermediate steps they need to achieve their end goal, or they may need to create an association between multiple, seemingly unrelated, events.
Although she’s recently been on sabbatical to focus on her research, McIlraith said that she always particularly enjoys teaching. “Maybe you write a fantastic paper, and what does that mean? Maybe a hundred people cite it, or a thousand people cite it… but you can really impact somebody’s life [by teaching],” she reflected.
Now that AI and machine learning are so prevalent in the public sphere, she says that it’s more important than ever to educate the general public — and AI-specialized computer scientists — about how to use it responsibly.
In fact, she’s been working with Dr. Barbara Grosz, the Higgins Professor of Natural Sciences at Harvard University’s John A. Paulson School of Engineering and Applied Sciences, to try to implement a new, specialized ethics course for the computer science department to tackle these issues.
She hopes that other AI researchers, like her fellow CIFAR AI chairs, will also push for a better public understanding of machine learning.
“[Machine learning] seems so magical, [but] people need to understand that these systems are only as good as their data,” she said.