The University of Toronto’s Student Newspaper Since 1880

Share on facebook
Share on twitter
Share on email

U of T AI Conference highlights use of machine learning to address the climate crisis

ProjectX winners forecast infection risk of fungal disease Black Sigatoka
Share on facebook
Share on twitter
Share on email
According to the co-presidents of U of T AI, going virtual allowed for a bigger conference with more high-profile speakers. COURTESY OF UOFT AI
According to the co-presidents of U of T AI, going virtual allowed for a bigger conference with more high-profile speakers. COURTESY OF UOFT AI

Artificial intelligence (AI) is one of the biggest fields of computer science research at the University of Toronto. One opportunity for students from various disciplines to get involved and showcase their own work is at the U of T AI Conference, which took place remotely on January 15 and 16. The conference aims to inspire students and researchers to use machine learning and AI to tackle the most urgent global issues — namely, the climate crisis. 

Organized by the U of T Artificial Intelligence student group (U of T AI), the conference featured renowned speakers from all over the world. Notable speakers included U of T Professor Emeritus Geoffrey Hinton, who is a pioneer in the field of machine learning, as well as other speakers from both large institutions and smaller entrepreneurial labs.

Winners of the ProjectX competition, a remote machine learning competition organized by U of T AI, were also announced during the conference.

 

AI for empowering the individual

Physicist Max Tegmark from the Massachusetts Institute of Technology spoke about using AI for sustainable science, development, and democracy. Using the United Nations’ sustainable development goals for reference, Tegmark remarked that most AI research is concentrated in areas with a lot of funds, such as health care and education in wealthy countries. In contrast, issues such as hunger and justice are largely ignored. 

In addition, Tegmark highlighted AI’s potential for both good and harm by discussing the inequities that it can help bridge or widen. For example, Tegmark referred to the influence of the media in creating what he calls “filter bubbles” in the increasingly polarized United States. Government or big tech fact-checking can easily lend itself to abuse, compromise freedom of speech, and be interpreted as censorship. However, machine learning can be inexpensively employed to reduce bias in news consumption by suggesting sources from both ends of the political spectrum.

Tegmark concluded that we need to think about “how [to] make an AI [so] that it doesn’t overpower us, but empower us, the individuals.”

AI and civil rights

Neuroscientist Vivienne Ming from Socos Labs, an entrepreneurial incubator based out of California, spoke about the importance of ethics in AI. According to her, while AI is often seen as a magical solution to our problems, we need to be able to understand and solve it first. Machine learning is based on history, which may not always be what we want to program. Ming used the failed Amazon hiring algorithm — which, until it was abandoned in 2018, only selected men as promising candidates — to highlight the importance of human intervention.

Ming also pointed out the increasingly invasive use of AI — such as in face recognition, job applications, and bank loan applications — and how easily it could be abused. She believes that we need “AI acting in our individual self-interest,” rather than in the interests of corporations.

Behind the scenes

For the first time, the U of T AI Conference was held entirely online. In an interview with The Varsity, organizers and co-presidents of U of T AI Shardul Bansal and Elias Williams discussed the theme of the conference and the advantages and disadvantages of the remote setting.

“We think [the climate crisis] is the most important [and] urgent problem that we have right now as a generation and… as a species,” said Williams. He further explained that “[the climate crisis] works well with machine learning… as [the field] is not as saturated as a lot of other machine learning disciplines are… and a lot of the data… is really readily available for free.”

Although there are new challenges associated with holding the conference online — especially with marketing and motivation — Bansal and Williams are certain that there are benefits as well. The online platform allowed them to reach a greater audience, broaden speaker options, and lower the cost for attendees. “We were able to have… all these people from really interesting places to talk because we didn’t have to fly them out,” Bansal concluded.

Project X

Project X is a remote machine learning competition organized by U of T AI that challenges undergraduate students all over the world to apply machine learning solutions to the impacts of the climate crisis. This year, teams came from not only Canada and the United States but also Germany, Poland, Chile, and Brazil. Project X is supported by a range of partners, including Google, Accuweather, IBM, and U of T’s Schwartz Reisman Institute for Technology and Society.

Over the course of three months, from September to November, 21 teams of undergraduate students competed in one of three focus areas to share in a $70,000 prize pool: infectious disease, weather and natural disaster prediction, or emissions and energy efficiency. This year, the winning teams were from U of T, University of Michigan: Ann Arbor, and Cornell University.

U of T winning team

The U of T winning team was a diverse group of undergraduates from across the Faculty of Arts & Science, spanning several time zones: Yuchen Wang, Matthieu Chan Chee, Ziyad Edher, Minh Duc Hoang, Shion Fujimori, Sornujah Kathirgamanathan, and mentor and co-author Jesse Bettencourt. 

Their project focused on infectious disease and used machine learning to develop a neural network to forecast the infection risk of Black Sigatoka, a fungal disease primarily attacking banana crops around the world. Unlike conventional models, the team’s neural network, Multiple predictoR Neural ODE, encodes external predictor factors such as weather conditions to improve the accuracy of the predictions.

“We first thought about modelling human diseases… [but] there are so many factors,” Chee explained in an interview with The Varsity. The group also had difficulties with finding data for the project as much of the data was privately owned. However, they succeeded with Black Sigatoka, and Kathirgamanathan felt that “[their project] has an even stronger connection between [the climate crisis and COVID-19].”

Their advice for aspiring students? Chee said, “There were many… big names [in this competition], and it’s easy for students to feel intimidated by this. But… don’t be afraid to try new things and go for it.” 

Team lead Yuchen Wang encouraged women to join and get involved. “Don’t be afraid to take the leadership role, and don’t be afraid to get involved in technology [or] academic research.”