Events

ConCats: Lifelong Concept Learning for Machines

Speaker: Mengye Ren

Location: Meyer Hall, 4 Washington Place, Room 575
Videoconference link: https://nyu.zoom.us/j/8169758973

Date: Friday, April 4, 2025

Today's large-scale AI models primarily acquire knowledge through offline, iid learning. While there is some in-context learning capability for online adaptation, it is crucial to let them acquire and consolidate new concepts at deployment. In this talk, I will introduce a few recent works on concept learning for foundation models, including learning new word concepts for LLMs and unsupervised in-context clustering. Lastly, I will also introduce an unsupervised event segmentation algorithm that can support lifelong visual representation learning.