Autocompletion of Design Data in Semantic Building Models using Link Prediction and Graph Neural Networks

Viktor Eisenstadt; Jessica Bielski; Christoph Langenhan; Klaus-Dieter Althoff; Andreas Dengel

In: Burak Pak; Gabriel Wurzer; Rudi Stouffs (Hrsg.). Co-creating the Future: Inclusion in and through Design - Proceedings of the 40th Conference on Education and Research in Computer Aided Architectural Design in Europe. Education and Research in Computer Aided Architectural Design in Europe (eCAADe-2022), CUMINCAD, 2022.


This paper presents an approach for AI-based autocompletion of graph-based spatial configurations using deep learning in the form of link prediction through graph neural networks. The main goal of the research presented is to estimate the probability of connections between the rooms of the spatial configuration graph at hand using the available semantic information. In the context of early design stages, deep learning-based prediction of spatial connections helps to make the design process more efficient and sustainable using the past experiences collected in a training dataset. Using the techniques of transfer learning, we adapted methods available in the modern graph-based deep learning frameworks in order to apply them for our autocompletion purposes to suggest possible further design steps. The results of training, testing, and evaluation showed very good results and justified application of these methods.


Weitere Links

ecaade_paper222.pdf (pdf, 488 KB )

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence