Recent advancements in artificial intelligence depend significantly on large-scale datasets with unified representations. However, such datasets are often unavailable for spatial computing tasks involving the generation and analysis of 3D data. In this talk, I will discuss several strategies for overcoming challenges posed by limited 3D data and outline how to develop AI systems with spatial intelligence. First, I will introduce how to design efficient generative modeling methods for particular 3D representations, such as point clouds. Next, I will demonstrate machine learning techniques to analyze geometric properties of 3D shapes, such as detecting partial and global symmetries. Finally, I will highlight future opportunities and outline ongoing challenges crucial to advancing spatial intelligence technologies.

Biography

Guandao Yang is a postdoctoral scholar at Stanford, where he works with Professor Leonidas Guibas and Professor Gordon Wetzstein. His research intersects computer graphics, computer vision, and machine learning. He completed his Ph.D. at Cornell in 2023 under the guidance of Professor Serge Belongie and Professor Bharath Hariharan. While pursuing his doctorate, he had experience collaborating with leading industry research labs including NVIDIA, Intel, and Google. Guandao earned his bachelor's degree in Mathematics and Computer Science from Cornell University. His research has received funding support from Magic Leap, Intel, Google, Nvidia, Samsung, and the Army Research Lab.