A Dilemma-Based Learning-to-Rank Approach for Generative Design in Urban Architectural Regeneration

Abstract

Continuous urbanization and climate change degrade urban living conditions. Nature-based solutions in architectural and urban design offer promising remedies but are often hindered by time, cost, and early design phase challenges. To address this, we present a Generative Design System framework utilizing AI-generated images and learning-to-rank algorithms. This system generates numerous image solutions to inspire architects and urban planners, significantly accelerating early design stages. To manage the overwhelming volume of images, we introduce a dilemma-based learning approach that employs learning-to-rank and smart bubble sorting algorithms to prioritize images based on user preference. A case study demonstrates the framework’s potential, providing valuable insights into its application, benefits, and limitations in urban design.