By Ian Gormely
April 13, 2022
The Vector Institute once again held its annual Research Symposium, showcasing the latest leading-edge research coming out of the Vector community. The annual event brings together Vector’s community of over 600 researchers, including some of the world-greatest minds in AI, to share their insights.
“Vector’s research symposium is notable because it features a lot of student and postdoc work,” says interim research director Graham Taylor. “With long timelines and the freedom to be curious, this kind of academic research can take larger risks that have the potential to pay off in big ways.”
He points to former postgraduate affiliate Patricia Thaine who presented her poster “Vec2int: Applications of the Chinese Remainder Theorem in Word Embedding Compression and Arithmetic,” at Vector’s Natural Language Processing Symposium in 2020. That research was the basis for her Toronto-based startup Private AI.
This year’s event included talks by Vector Faculty including Angela Schoellig and Juan Felipe Carrasquilla and poster presentations from members of Vector’s community of over 600 researchers. Faculty members who attended this year’s two-day event were impressed by the research results that showcased the range of ways that AI could help improve society.
Vector researchers enjoy academic freedom to pursue any potential line of inquiry, though that work is often focused on addressing fundamental questions in AI, particularly in health care, sequential decision making, generative models, machine learning, AI theory, and security, privacy, and fairness. Their work has the potential to deepen our understanding of AI, lead to applications that improve industry and quality of life, and push forward the field of responsible AI.
“Learning to Elect,” was a particular highlight for many. Co-authored by Vector PhD students Cem Anil and Xuchan Bao, the paper details how a neural network could be used to discover the most suitable voting rules for a given situation. “The paper came from a project for the amazing course “Algorithms for Collective Decision Making” taught by Vector Faculty Affiliate Nisarg Shah,” says Bao. Voting is often used in the hope of maximizing some notion of social welfare, but voting rules don’t always match up with this goal.
“As a learning task, it’s very beautiful and potentially very useful for society,” says Juan Felipe Carrasquilla, a Vector Faculty member. “You can imagine applying it to many different things, not just elections.”
Taylor also highlighted Parand Alizaadeh, Toryn Klassen, Rodrigo Toro Icarte, and Sheila McIlraith’s paper “Avoiding Negative Side Effects by Considering Others” in which a reinforcement learning agent was given the ability to consider the impact of its actions on the wellbeing and agency of others in the environment, like other humans.
Both Taylor and Carrasquilla underscored the important role events like the Research Symposium play in community building. “Given the early-stage nature of a lot of the research, this is a point at which you may have questions or suggestions as a poster visitor and the person at the poster says, ‘Thank you for that suggestion, I am excited to try it out,’” says Taylor. Bao notes that they received “very fruitful feedback” on their paper at the Symposium.
“Especially with the pandemic, we’re so isolated,” says Carrasquilla. “It brings us together through research and can spark collaborations for future projects.”
Want to learn more about Vector researchers’ work? Sign-up for our newsletter here.
Looking to join our research community and are affiliated with a university from Vector’s network?