Newest and Lowest Rated Papers
Advanced filtersTopic: AllExperimental
Topic
Topics are auto-detected from title, abstract, and metadata and may be imperfect.
Expert Evaluation of a House Floor Plan Recommendation System Using Graph Neural Networks
Hyejin Park, Mini Heo, Yeji Sẹo, David Stephan Panya, Hyeongmo Gu, Seungyeon Choo
In the pre-design stage of architectural planning, reference research is often conducted
manually through books, magazines, and real estate websites. Websites like Archdaily and
Dezeen filter data based on input criteria such as the number of levels, rooms, or
approximate building area. However, this approach is time-consuming as users must
manually review and select information, making it nearly impossible to find floor plans
based on spatial conditions. In this study, the pre-design stage refers to the early phase
where spatial requirements and reference ideas are explored before schematic design
begins. For houses, the diversity in personal preferences and styles often results in
frequent changes to client requirements, prolonging discussions and delaying immediate
responses. To address these challenges, the House Floor Plan Recommendation System
(referred to as AIBIM-House Reference Finder) was developed in prior research. This
system collected approximately 10,000 raster-based house floor plan images
(representing around 6,900 buildings) and utilized the You Only Look Once (YOLO)
model to create a vector-based spatial relationship dataset. The dataset was trained using
Graph Neural Networks (GNN) to develop a similarity-based recommendation system
considering spatial shape, adjacency, and relationships. While the system’s quantitative
performance was validated earlier, this paper evaluates its qualitative utility through
expert evaluation. Architectural design experts assessed the similarity of recommended
floor plans and the system's usability in the pre-design stage. Similarity evaluations
identified key factors, including spatial shape, adjacency, direction, size, and number, as
primary criteria for judging similarity. Usability evaluations highlighted the system's
effectiveness in zoning, drawing, client discussions, space programming, and feasibility
analysis. This paper demonstrates the potential of integrating AI-based spatial conditions
into traditional reference research methods, automating alternative exploration and
generation processes. This approach enhances exploratory creativity, advances
transformative creativity, and improves the efficiency of planning tasks.
No ratings yet
View paper →A Design Theory for Fast and Slow Thinking in Interactive Coding
Evan S. Raskob
This paper introduces a new theory and shared language to help users and designers of interactive
programming (IP) systems, especially those aimed at producing “creative” multimedia outcomes such as
generative sculpture, art, music, graphics, and livecoding performances, to better identify key
programming activities and be more strategic about their application. The theory was inspired by two
complementary modes of thinking, “fast” (intuition) versus “slow” (deliberation). It defines three key
“activity clusters” called LiveCoding Performance, Experimental Interactive Programming, and
Exploratory Interactive Programming using the Cognitive Dimensions of Notations as a language to
discuss practice-based reflections and user studies of intuitive and deliberative activities in
performances, structured experimentation and software development, and open-ended creative
exploration. It interrogates the usability trade-offs between the three clusters during the design of
LivePrinter, the author’s interactive 3D printing (I3DP) and performance system. This positions these
activities within the author’s practice as an educator, sculptor, performer, and software developer who
experiments, composes, and performs, leading to some guidelines for other “creative” IP developers.
No ratings yet
View paper →