Cover Image

Digital intuition – Autonomous classifiers for spatial analysis and empirical design

Christian Derix, Prarthana Jagannath

Abstract


Unsupervised network models have found application as data mining tools in spatial analysis and as search tools in spatial design. Building on previous work at the urban scale, ‘self-organising feature maps’ (SOMs) are tested as mechanisms to classify heterogeneous spatial attributes of layouts and extract representative data profiles. These profiles have been presented in the paper as potentially new spatial typologies, which are implicit within the morphology of a layout and are representative of the occupational affordances of that layout. Contrary to traditional typologies, these typologies are derived rather than assigned. The methods explored in this paper establish a way in which spatial typologies can be autonomously recognised through associations between spatial attributes of a layout. Spatial typologies which are representative of occupant perception may be useful in incorporating user-centric approaches more systematically into the design process and could give rise to new key-performance indicators.

Although the models were developed and are described using test cases established in the context of spatial resilience and infrastructure security, they are presented with the intention of generalisation and with an eye on their application to spatial analysis and design outside the remit of security. The paper aims to focus on the development and structure of these specific models which are later generalised beyond the security perspective for application in other design sectors. The models are intended to act as heuristic underlays for the design process, maintaining light-weight interactivity and real-time feedback. This is largely to introduce analysis earlier in the design process and enable an iterative design-analysis workflow. The paper also speculates on the future of design methodologies beyond parametric procedures.

Keywords


Spatial typologies, artificial neural networks, self-organising maps, associative models, spatial characterisation

Full Text: JOSS_2014_P190-215