Generative AI technologies have proved their usefulness in several information processing applications, but struggle with complex tasks such as spatial reasoning and complex design problems. As a result, they are inadequate for a truly transformative impact in the design and manufacturing domain due to a variety of shortcomings, including those With creating unrealistic 3D models. We combine deep learning with the symbolic graph grammar framework to tackle the inverse design problem for 3D shapes. We develop a versatile infrastructure for this system, including building and pre-processing a high-quality, comprehensive 3D shape dataset, advancing grammar learning algorithms, and refining 3D shape generation and analysis methods, to produce geometry and topology-aware 3D shapes.

I am participating in SuperUROP because I would like more experience in research. I worked on computational geometry research projects in high school and would love to delve deeper into this field. I believe 3D shapes are very important for a lot of growing fields, such as VR and robotics, so I’m excited to work on cutting edge techniques to generate accurate 3D models.