2024
Courtney Paquette
Mathematician followed unusual path to machine learning research
2024
Mathematician followed unusual path to machine learning research
Courtney Paquette, Assistant Professor of Mathematics at McGill University, didn’t foresee a career in machine learning when she was doing her undergraduate degree in finance. At the time, she was studying supply chain management and learning how to use mathematics for efficiencies in ordering supplies, which introduced her to the field of optimization probability.
“I really liked it,” she recalls, which led her to taking additional courses in this area. “While I was in a numerical analysis course, the professor basically said, ‘What are you doing? You’re really good at math. Do you want to do a PhD in math?’”
That’s when she pivoted in her studies to optimization probability — and found she had some catching up to do. “It was hard because I didn’t actually have the background [for a PhD],” she says.
A year spent working at what is now known as Google DeepMind led Dr. Paquette to focus on machine learning “because that’s where optimization was going at the time.”
“At Google, I got to interact a lot with computer scientists (I’m more on the theory side). They introduced me to some cool problems that they were asking about theoretically, which had very applied applications, and that’s how I ended up working on the practical implications of machine learning.”
Among other things, Dr. Paquette — a Canada CIFAR AI Chair — is working to make machine learning algorithms more efficient in terms of compute (time and cost) budgets.
“The problem with running these really big models, like large language models [such as those used by Chat GPT], is that they’re so big and we can’t run them for very long; we’re limited by compute. So we need to be able to sort of predict the behavior of an algorithm so that it will run in the most compute-efficient way.”
Dr. Paquette has been named a 2024 Sloan Research Fellow. She welcomes the award as recognition of the value of her research.
“I’ve been really trying to change the way we do optimization, so it feels nice to be recognized for trying to do something a little bit different from what is normally done.”
