Classifying building and ground relationships using unsupervised graph-level representation learning

Alymani, Abdulrahman, Mujica, Andrea, Jabi, Wassim ORCID: https://orcid.org/0000-0002-2594-9568 and Corcoran, Padraig ORCID: https://orcid.org/0000-0001-9731-3385 2022. Classifying building and ground relationships using unsupervised graph-level representation learning. Presented at: Design Computing and Cognition’22, Glasgow, Scotland, UK, 4-6 July 2022. Published in: Gero, John S. ed. Design Computing and Cognition’22. Springer: Springer Nature, pp. 305-320. 10.1007/978-3-031-20418-0_19

Preview

PDF - Published Version Download (712kB) | Preview

Abstract

When designing, architects must always consider the ground on which their buildings are supported. Our aim is to use data mining and artificial intelligence (AI) techniques to help architects identify emerging patterns and trends in building design and suggest relevant precedents. Our paper proposes a novel approach to unsupervised building design representation learning that embeds a building design graph in a vector space whereby similar graphs have comparable vectors or representations. These learned representations of building design graphs can, in turn, act as input to downstream tasks, such as building design clustering and classification. Two primary technologies are used in the paper. First, is a software library that enhances the representation of 3D models through non-manifold topology entitled Topologic. Second, an unsupervised graph-level-representation learning method is entitled InfoGraph. Result experiments with unsupervised graph-level representation learning demonstrates high accuracy on the downstream task of graph classification using the learned representation.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Publication
Status:	Published
Schools:	Architecture
Publisher:	Springer Nature
ISBN:	9783031204180
Date of First Compliant Deposit:	5 July 2022
Last Modified:	28 Apr 2024 17:53
URI:	https://orca.cardiff.ac.uk/id/eprint/151063

Actions (repository staff only)

Edit Item