Deciphering the noisy landscape: Architectural conceptual design space interpretation using disentangled representation learning

Abstract

Time and resource restrictions limit the architect's design scope. Computational design methods can offer support to overcome these limitations. Design exploration has been a long‐established task in computational‐aided generative design. However, conventional objective‐ and performance‐based systems have restrictions pertaining to the exploration scope. Without a quasi‐global cognition of the conceptual design space, the exploration scope is bound to be limited. This paper is a proposal for an epistemic shift toward the interpretation of conceptual design space per se. This topic receives limited attention in the current literature due to the scarcity of interpretation tools. Using a customized large‐scale architectural image database with high‐level visual diversity and latent data space coverage, this paper serves as a first attempt to investigate the possibilities of leveraging disentangled representation learning to structurally interpret architectural conceptual design space in both supervised and unsupervised manner. Various schemes of supervised disentanglement are tested, with analytical comparisons indicating discrepant structural traits of different latent spaces. The unsupervised interpretation scheme shows the preliminary capability of automatic feature disentanglement. Our long‐term objective is to offer designers a broader spectrum of creative design through innovative design systems.

Application of multimodal speech recognition based on deep neural networks in interpretation teaching

“…It is conveyed to the audience through specific intonation and translation, using words, symbols, or sound forms as carriers. In the process of speech output, the interpreter bears the psychological transformation of understanding the content of the text [19][20] . Therefore, only when the interpreter is able to promptly correct and ensure accurate transmission of the translation when making mistakes can the expected purpose be achieved.…”

From Exploration to Interpretation - Adopting Deep Representation Learning Models to Latent Space Interpretation of Architectural Design Alternatives

“…Neural network-based graph generation is a burgeoning field explored in various domains like molecules, protein structures, and scene graphs [12], [14]- [16]. However, in architectural design research, while many studies focus on generating architectural data in Euclidean formats like images and 3D models [8]- [10], the specific task of generating architectural layout design graphs has been notably unexplored.…”

“…Previous studies primarily focused on superior exploration strategies, leaving the intrinsic structure of design representation space vague [4]. As design activities are made possible because of designers' mental models of design representation spaces that designers constantly perceive and formulate [4]- [6], Chen & Stouffs [6], [8] promote two explicit models of design representation spaces: the sparse humanlearned model and the compressed machine-learned model, arguing that designers may enhance design performance by interacting with simulated design representation spaces. In this context, converting architectural design data into machineinterpretable formats is necessary, requiring flexible representation learning schemes.…”

“…In this context, converting architectural design data into machineinterpretable formats is necessary, requiring flexible representation learning schemes. Recent data-driven techniques have demonstrated the ability to interpret design concepts by converting data into vectors of neural activities [8], [33]. However, current studies on architectural design representation focus mainly on Euclidean data, with limited exploration of non-Euclidean data like graphs.…”

“…This suitability extends from macro-level layouts to micro-level construction details, allowing for an interconnected view of design elements and illustrating their cohesive functioning. Architectural design space serves as a fundamental concept in design research [3]- [7], yet the representation of architectural design space and interpretation of corresponding design representation space using deep learning-based approaches [8], especially concerning structured non-Euclidean data like graphs, remains understudied. Most existing studies focus on Euclidean design data formats [8]- [10], leaving a gap in exploring graph-based representation learning and synthesis in architectural design.…”