As Generative AI (GenAI) becomes a more prominent topic in our society, it is no surprise that there is growing interest in exploring this technology in the research sector.
At the U of C, Associate Professor and research chair Dr. Yani Ioannou and his team are working to improve these AI models in multiple ways.
Ioannou spoke to the Gauntlet about his research and how the rise of AI provides new opportunities for engineering students.
To many, GenAI is a frontier of endless possibility. In reality, there are many issues connected to many of the major AI models, such as concerns about efficiency, transparency, accuracy and reliability. Ioannou works to find solutions to these problems by developing new methods to refine how models are run or trained.
For a majority of individuals, using advanced AI models is not a matter of “how” but a matter of “how much.” Companies often charge individuals for the ability to run the high-end versions of AI tools through subscriptions. There is an even greater cost in training better-quality models, as it often requires massive computational cost. This results in a significant price barrier for accessing the most sophisticated AI.
“So right now, it's very much big tech that has access to that technology or potentially very large organizations. And that inevitably means that, you know, rich countries, by default, have more access to it,” said Ioannou
The professor's current research centers on democratizing AI, or making AI tools and resources equally available for all. To support his work, he has received funding from NSERC in collaboration with IVADO to work on research with the French National Research Agency (ANR).
Ioannou is a co-PI on one of his projects with Dr. Umut Şimşekli of the French National Institute for Research in Digital Science and Technology (Inria), one of the country’s top research institutions. The pair specifically work with neural networks, AI models that process information in a manner inspired by human neurons. They blend their different expertise to find ways to use fewer computational resources for high-performance models.
“[Şimşekli] is a machine learning researcher, but he is that much on the math side of things, whereas I'm a much more empirical person, working on actually running experiments, training, and evaluating models,” said Ioannou. “We wanted to put those skills together and try to understand if we can make these neural networks more accessible.”
One method for expanding the availability of AI models is through specialization. Ioannou describes how AI hobbyists have found ways to modify models for specific tasks, which can reduce the size and resource cost of these models.
“Often when we use these models, it's for one topic like coding or creative writing, or gardening,” said Ioannou. “You're not typically likely to change from gardening to coding within the same set. So you can specialize those models and make them much smaller for those specific tasks.”
Ioannou compares this to the evolution of earlier technology. Similar to how the hobbyist movement of the 1970s led to increased accessibility of computers, today’s enthusiasts could be aiding in a similar progression for AI.
Ioannou is the lead of the Calgary Machine Learning Lab, a research group at the U of C where students work with AI and optimize how models can be used. For graduate and PhD students, joining the lab provides an opportunity to become mentors, work on longer projects and collaborate with major tech companies. Most undergraduate students work in the lab for a few months at a time during summer studentships, funded by awards such as the Schulich 50:50 or Program for Undergraduate Research Experience (PURE). These opportunities allow students only a brief time to work on a research project; Ioannou still encourages students to apply to these positions, emphasizing how the mentorship and learning aspect of these programs is a large part of their significance for undergraduates.
“Our job is to do research, but it is also to train [students],” said Ioannou. “It’s very important to make progress on the research side, but also to give [Canadian students] the expertise in this area.”
All across the world, enrollment in undergraduate and graduate Computer Science and Software Engineering programs has been on the decline. Ioannou theorizes that this is due to many students believing that technology has advanced to the point where significant human contributions can no longer be made. He argues that the emergence of AI is not limiting innovation but creating new paths for exploration.
There is a common yet false assumption that AI is the pinnacle of technological advancement; infallible and unflawed. Professor Ioannou’s research illustrates how there is still much room for improvement in this technology and ways for students to contribute to this change.
