Top

U of T grad Sara Sabour helps computers see the world more like a human

Sara Sabour (Photo: provided)

As she prepares to graduate with a PhD in computer science from the University of Toronto, Sara Sabour reflects on a journey that has taken her from high school programming competitions to cutting-edge research at Google and collaborations with AI pioneer Geoffrey Hinton. Her work in computer vision and machine learning, particularly in object representation and 3D scene understanding, has contributed to making artificial intelligence systems more robust, interpretable and human-like in their perception of the world. 

In this conversation, Sabour, who previously earned a Master of Science degree in computer science from U of T in 2016, shares insights into her research, the challenges of doctoral studies and the mentors and moments that shaped her path. 

This interview has been edited for clarity and length. 

What first drew you to the field of computer science (CS)? 

In high school, I competed in programming Olympiads and realized I was very good at solving problems and coming up with algorithms. I could spend hours coding without getting tired. That passion led me to study computer science in undergrad. Initially, I leaned toward theoretical CS, but a machine learning course opened my eyes to how computers could learn on their own. That fascination eventually led me to computer vision and a PhD.

After you completed a master’s degree in CS from U of T in 2016, you joined Google as a researcher and worked with Geoffrey Hinton there. How did that experience shape your academic path? 

Working with Geoff reignited my passion for research. His joy in solving problems, regardless of deadlines or recognition, was inspiring. That experience convinced me to pursue a PhD. I later joined the CS PhD program at U of T, where Geoff co-supervised my work alongside David Fleet. Both were incredibly supportive and influential in shaping my research direction.

Tell us more about your PhD research. 

My work focuses on teaching AI to understand objects in images similar to how humans do. Traditional models often rely on shortcuts, like associating green backgrounds with dogs and blue backgrounds with fish, which makes them fragile. I worked on capsule networks and unsupervised learning methods that help models recognize objects based on their parts and spatial relationships. This makes them more robust and better at generalizing to new situations, like recognizing a face from different angles or in unfamiliar settings.

What are some real-world applications of your research? 

One example is smart home cameras. Current models might misidentify a spider on the lens as a package. Our methods can better detect and remove unexpected objects. In autonomous driving, this research could help identify unusual signs or obstacles. I’m also applying this to camera pose estimation — figuring out where a camera is in space — which is useful for VR and for training generative AI models that can simulate 3D environments.

What challenges did you face during your PhD, and how did you manage them? 

The most challenging aspect for me personally was the flow of work during the PhD. The workload often came in waves — quiet months followed by intense periods with deadlines, reviews and exams all at once. I learned to focus on one task at a time and accept that it’s okay not to juggle everything perfectly. That mindset helped me stay grounded.

How did your time at U of T shape you personally and academically? 

The diversity at U of T was eye-opening. I met people from all over the world with unique approaches to life and work. It helped me grow more open-minded and appreciative of a wide range of perspectives.

Who were your biggest mentors during your studies? 

Both Geoff Hinton and David Fleet were incredible mentors. David was very supportive and open to students exploring outside of their very specialized areas of research. Geoff’s passion was contagious. He’d stay up all night running experiments just out of curiosity. They both taught me that genuine interest in a problem is the best motivation.

What are you most proud of from your PhD experience? 

One highlight was mentoring high school students, especially girls in STEM. Some had internalized the idea that computer science wasn’t for them. I was proud to help challenge those assumptions and show them that success in this field doesn’t require fitting a certain mould.

What advice would you give to future CS PhD students at U of T? 

Take full advantage of the learning opportunities a PhD offers. I’m glad I took courses outside my department that challenged me and broadened my perspective. Also, don’t hesitate to reach out to professors beyond your supervisor. Many are open to collaboration and happy to support students who show initiative.

What’s next for you after graduation? 

I’m taking a short break, then continuing some projects at Google. I want to scale up my research so it can have a broader impact and help push the field forward. I’m also exploring academic positions, especially because I’ve enjoyed mentoring and teaching.