By: Erin Vollick
9 Oct, 2023
Adam Oberman, Mila: Leveraging math to address AI reliability
October 9, 2023
Adam Oberman, Canada CIFAR AI Chair, Mila
Generative AI models such as ChatGPT learn from large data sets with hopes that the model will choose the most efficient outcome — with unreliable results. That’s where Adam Oberman, Canada CIFAR AI Chair, has been making an impact. Concentrating on a form of deep machine learning that involves neural ordinary differential equations, Oberman creates faster, more reliable outcomes for AI agents. As one example, he’s found that asking the program to choose the shortest path to reach its conclusion can make training these types of models 10 times faster.
“My collaborations among the CIFAR community are already providing valuable research and helping me recruit trainees who bring fresh perspectives to the challenges we’re addressing.”
Adam Oberman, Canada CIFAR AI Chair, Mila
Oberman’s foundational principles are being widely implemented: former trainee Chris Finlay is building on this work at AI startup Deep Render, which hopes to revolutionize video compression. Meta AI and other companies also draw on these programs to artificially generate detailed images.
In another project, Oberman has applied geometric ideas to address bias in facial recognition programs. Oberman hopes that the code will contribute to ending discrimination in face verification AI — along with its profound effects on individuals.
“Being a member of this really impressive group of Canada CIFAR AI Chairs has gotten me excited. It’s been like a new career for me,” says Oberman. “My collaborations among the CIFAR community are already providing valuable research and helping me recruit trainees who bring fresh perspectives to the challenges we’re addressing.”