Autodesk Research’s Post

View organization page for Autodesk Research, graphic

7,049 followers

In engineering, generative models often face two main issues: obtaining diverse, high-performing datasets and ensuring generated designs meet precise constraints. This paper proposes an innovative approach for architectural design that combines optimization, constraint satisfaction, and language models. Autodesk Research used a method called Quality-Diversity (QD) to create a varied, high-performing dataset. Researchers then fine-tuned a language model with this dataset to create high-level designs, which are refined using the Wave Function Collapse algorithm to create detailed layouts that meet design constraints. Learn more here: https://bit.ly/45xuv6J

  • No alternative text description for this image
M. Kyan B.

Working at the intersection of: 🤖 Graph ML/Deep Learning || 👀 Computer Vision || 🏛 Built Environment/Heritage || 🏗️ Construction Management (with BIM)

2mo
Like
Reply

To view or add a comment, sign in

Explore topics