Events

CILVR Seminar: Continual learning and meta-learning in the era of foundation models

Speaker: Dr. Sung Ju Hwang

Location: 60 Fifth Avenue, Room 7th floor open space

Date: Wednesday, October 30, 2024

Continual learning, which involves learning over a sequence of domains or tasks, and meta-learning, which aims to acquire general knowledge across them for better generalization to unseen domains/tasks, have been important topics in machine learning research. However, with the rise of large-scale foundation models, such as language models with billions of parameters with no clear boundaries across tasks, these paradigms have received relatively less attention, as conventional approaches are often incompatible with, or nearly infeasible for, such models. Despite the rise of foundation models, the challenges addressed by continual and meta-learning remain relevant, and we simply need new approaches that are compatible with these models. In this talk, I will introduce some of the methods we have developed in this context, with the focus on multi-agent, on-device systems for embodied agents.