Mila Core Academic Member Sarath Chandar was appointed Canada Research Chair in Lifelong Machine Learning on August 29, 2023, and received a grant from the Canadian government to further his research work.
Sarath Chandar, also, Assistant Professor at Polytechnique Montréal and Canada CIFAR AI Chair, leads a research group focused on granting AI systems the ability to learn and improve continually over the course of their lifetime. His research spans across deep learning, optimization, reinforcement learning, and natural language processing.
“The ability to adapt quickly to new problems in an ever-changing world is an essential part of human intelligence. Humans achieve this by accumulating knowledge throughout their lives and reusing it to accomplish new tasks,” Professor Chandar explained.
“Although machine learning research has made significant progress thanks to advances in the field of deep learning, adapting to changes in the world remains a significant limitation,” he added.
The Canada Research Chair he leads seeks to remedy this limitation by exploring the paradigm of lifelong learning.
Continual learning systems, which assimilate multiple tasks over time, should be able to transfer previously acquired knowledge to new tasks in order to learn how to solve them efficiently.
However, several challenges need to be addressed before enabling continual learning capabilities in artificial neural networks.
For example, machine learning systems often tend to “forget” previously learned tasks while learning a new task, called “catastrophic forgetting.”
They also lose their ability to learn after some point, known as “loss of plasticity.”
Most modern AI systems lack the long-term memory needed to develop knowledge that can be reused.
Professor Chandar’s research focuses on tackling all these problems that arise while developing a lifelong learning system.
“I am honoured by this Canada Research Chair appointment, which will allow me to continue my ongoing research projects and create new ones by recruiting more students in my research lab. I hope to make a lasting impact in the field of lifelong learning agents and help make AI tools more performant and trustworthy,” Professor Chandar said.
To promote research in lifelong machine learning, he recently co-created the international Conference on Lifelong Learning Agents (CoLLAs), bringing together researchers whose goal is to improve the versatility of AI systems by reimagining paradigms of traditional machine learning systems.
Photo Credit: Caroline Perron