Composing Frankensteins:Data-Driven Design Assemblies through Graph-Based Deep Neural Networks

Over the last five years, machine learning and AI became exceedingly popular due to significant developments in the study of deep neural networks (DNN) or deep learning.Current research on DNNs focuses mainly on image and audio-based applications, ranging from self-driving cars, to virtual assistants such as Siri, to all kinds of online recommendation systems. Example applications include image classification, prediction of user choices, and generation of new images. In this paper, we present an alternative graphbased DNN approach to generate new conceptual designs. We trained DNNs with residential design represented through attributed graphs. We discovered essential building blocks based on performance criteria and composed them into new user-desired assemblies – aided by learned information about the proximity of various design components in latent vector space.

https://doi.org/10.35483/ACSA.AM.107.90

Volume Editors
Amy Kulper, Grace La & Jeremy Ficca

ISBN
978-1-944214-21-0