About.      Work.      


A voxelized plot of a knot surface defined by signed distance function



knotGAN
2020 -2021 (In-Progress)

Design Software
GH Component - C#, Python
Interface Design

Team: Prof. Andrew Witt, Gia Jung, Claire Djang, Eunu Kim


The project is a continued investigation into creative potential and architectural application of 3D shape representation learning. In the mathematical field of knot theory, a knot invariant is a quantity (in a broad sense) defined for each knot which is the same for equivalent knots. The equivalence is often given by ambient isotopy but can be given by homeomorphism. As analytically indescribable, the research investigates deep-learning-based methods to describe the non-orientable surfaces.



Final Trained Model & Data Exploration from the knotGAN Tool






Generative Design Tool (GH Component) using 3D Shape GAN




Self-Organizing Map of Any Given 3D Dataset



Enumerative process to generate knot surfaces



Knot Surface Data:


Knot Surface Characteristics: Diagram



Generated Data Example


Generated Data Example

Subset of the Latent Space of Generated Knot Surfaces


Tool in Development To view Original Knot Surface vs. Generated Results
As Knot surfaces contain much potential in architectural application, we hope to further develop metrics to enhance topological classification of the hybrid knot surface.  


48 Quincy St. Cambridge, MA 02138