Content

Speaker

Zitian Chen

Abstract

AI systems increasingly face the challenge of understanding and integrating diverse types of information to navigate the real world better. A self-driving AI, for instance, must simultaneously detect lanes, track humans and vehicles, recognize traffic signs, and more—each task contributing to the overall system's functionality. In this talk, I will explore neural network architectures that facilitate joint learning across varied tasks. I will begin by presenting the key motivation behind this challenge, followed by a review of my work, which progressively tackles more complex and larger-scale scenarios. These efforts not only unify multiple vision tasks but also explore scalable solutions, further advancing our understanding of scaling laws in multi-task models.

Advisor

Erik Learned-Miller

Hybrid event posted in PhD Thesis Defense