By: Erin Vollick
10 Oct, 2023
A stratospheric balloon floats 20 kilometres above the earth, bringing internet connections to Kenyan communities. The balloon decides its own flight pattern in an energy-efficient way and – based on its interactions with the environment – discovers new, even more efficient behaviours. This was a live pilot project that Marlos C. Machado, Canada CIFAR AI Chair at Amii, collaborated on while working at Google Brain, alongside Marc Bellemare, also a Canada CIFAR AI Chair at Mila. The resulting reinforcement learning AI agent helps the balloon navigate the complexities of its environment. The work led to a high-impact Nature publication and the technology could be adapted for balloons that track and forecast weather patterns, forest fires, and more.
“Without CIFAR, I would be doing research for a U.S. company and not benefiting Canada as much as I could by training students and performing this research.”
Marlos C. Machado, Canada CIFAR AI Chair, Amii
Currently, Machado is collaborating with fellow Amii Canada CIFAR AI Chairs Martha White and Adam White on continual learning systems for AI agents. The team is finalizing a model for a water treatment facility that they hope to deploy live, in which continual adjustments are made by an AI agent to optimize the plant’s operations, 24 hours a day. Machado sees future applications for this research such as smart housing that can lower emissions while optimizing human health.
“CIFAR funding allowed me to do this research with students,” says Machado, who will be offering a public lecture on the balloon’s AI system for event series Jasper Dark Sky, part of a partnership between Amii and Tourism Jasper, this autumn. “Without CIFAR, I would be doing research for a U.S. company and not benefiting Canada as much as I could by training students and performing this research.”