This will be a hybrid event with in-person attendance in Levine 307 and virtual attendance on Zoom.

ABSTRACT

Modern artificial intelligences (AIs) rely heavily on internet-scale data with unified representations. However, such large-scale homogeneous data isn’t readily available for spatial computing applications involving 3D geometry, hindering the development of spatial intelligence— AIs that can generate and understand 3D spatial data. In this talk, I will present ideas toward building spatial intelligence systems with limited 3D data. I will discuss my work combining existing mathematical models in graphics with foundation models in machine learning to generate and analyze 3D shapes. Finally, I will conclude with a discussion about the future opportunities and challenges in developing data-efficient AIs for spatial computing and beyond.