1 h 5 min
Seminar, Hosted by Alberta Machine Intelligence Institute
Scaling, Robustness and Continual Learning in Foundation Models: Recent Advances from the Autonomous AI Lab
Université de Montréal and Mila – Quebec AI Institute
Recording
Abstract
Irina Rish surveys how foundation models scale, respond to shifts in their inputs and adapt without being retrained from scratch. The talk covers contextual time-series forecasting, efficient model architectures, preservation of intermediate representations for reasoning, out-of-distribution evaluation and methods for continual pretraining and compression.