Sequencing the Architectural Design Process for Artificial Intelligence: A design-theory-based framework for machine learning approaches

Similarprocess models of the architectural designprocess of the early design stages have beenformalised. However, recognition by machine learning (ML) based approachesfails due to the individuality and ^vagueness of the inherent method of sketching. Nevertheless, contemporary ML approaches have the potential to support the architectural design process through auto-completion-based suggestions. In order to provide data for ML- based suggestion generation, wepropose a customisablejramework with according steps. Drawingfrom design theory, it is establishes the designprocess as sequences of three levels of detail and their respective linking. These literature-based sequences serve to label sketch protocol studies. Finally, theframework is validated through Recurrent Neural Networks (RNNs) with Long-Short-Term-Memory (LSTM) architecture trained in isolation on sequences of different level of detail, for prediction purposes.