Dimensionality reduction for parametric design exploration

Harding, John

Abstract

In architectural design, parametric models often include numeric parameters that can be adjusted to explore different design options. The resulting design space can be easily displayed to the user if the number of parameters is low, for example using a simple two or three-dimensional plot. However, visualising the design space of models defined by multiple parameters is not straightforward. In this paper it is shown how dimensionality reduction can assist in this task whilst retaining associativity between input designs in a high-dimensional parameter space. A form of dimensionality reduction based on neural networks, the Self-Organising Map (SOM) is used in combination with Rhino Grasshopper to demonstrate the approach and its potential benefits for human/machine design exploration.

Citation

Harding, J. (2016). Dimensionality reduction for parametric design exploration. In S. Adriaenssens, F. Gramazio, M. Kohler, A. Menges, & M. Pauly (Eds.), Advances in Architectural Geometry 2016 (274-287). Zurich, Switzerland: vdf Hochschulverlag AG an der ETH Zurich

• Conference Name: Advances in Architectural Geometry 2016
• Conference Location: Zurich, Switzerland
• Acceptance Date: May 18, 2016
• Publication Date: Jan 1, 2016
• Deposit Date: Oct 10, 2016
• Publicly Available Date: Oct 20, 2016
• Peer Reviewed: Peer Reviewed
• Pages: 274-287
• Book Title: Advances in Architectural Geometry 2016
• ISBN: 9783728137777
• Keywords: parametric design, machine learning, dimensionality reduction, self-organising maps, data visualisation
• Public URL: https://uwe-repository.worktribe.com/output/1435025
• Publisher URL: http://vdf.ch/advances-in-architectural-geometry-2016.html
• Additional Information: Additional Information : This work is licensed under creative commons license CC BY-NC-ND 2.5 CH (https://creativecommons.org/licenses/by-nc-nd/2.5/ch/); Title of Conference or Conference Proceedings : Advances in Architectural Geometry 2016