What does the future of AI informed design look like? Recent advancements in artificial intelligence and deep learning have created opportunities for design experimentation through the use of neural networks. This summer, I had the privilege of working with Professor Andrew Saunders in the Architecture Department at the Weitzman School of Design to investigate how deep learning neural networks can influence the design process. This project experimented with convolutional neural networks (CNN) to generate 2D visual compositions that were then used to inform 3D architectural designs.
To create AI generated images I learned how to input style and content images into a CNN. By experimenting with the CNN I developed an understanding of how the code functioned in order to produce new images. Following this process, I collaborated with my fellow research assistants to produce a 3D modeled pool based on a composition produced by CNN. Although we used an AI generated image to produce the pool, we also had to make our own creative decisions when determining how to interpret different aspects of the image. In addition to creating designs based on CNN images, I also assisted on Professor Saunders’s own project based on deep learning neural networks. Curated by Penn faculty members Andrew Saunders, Robert Stuart-Smith, and Scott Erdy, this installation will be a robotic relief project at a Community Center and Library.
As a design student pursuing a minor in architecture, this project introduced me to new ways of approaching design decisions. In addition, I was able to get hands-on experience working in a field that I have always been interested in. Not only did I have the opportunity to make my own creative decisions, I contributed to a design project that will be installed in a public space and have real-world impacts. Overall, my work with Professor Saunders gave me valuable insight into my desired career path as well as experience working in an actual design lab.
To see my poster, visit Penn Presents: https://presentations.curf.upenn.edu/poster/using-neural-networks-inform...