LayoutGKN: Graph Similarity Learning of Floor Plans

2025-09-03

AI Summary

Existing graph matching networks for architectural floor plan (graph-structured) similarity computation suffer from low inference efficiency due to early cross-graph node interaction. Method: This paper proposes a delayed-interaction framework for efficient graph similarity learning. It defers cross-graph interaction to the final node embedding layer and employs a differentiable graph kernel as the distance function, enabling direct graph-level similarity computation in the final embedding space—bypassing costly intermediate node alignment. The method builds a joint embedding architecture based on graph neural networks and optimizes the graph kernel distance end-to-end. Contribution/Results: Experiments on multiple architectural graph datasets show that the proposed method maintains or even surpasses the accuracy of state-of-the-art graph matching networks while achieving significantly faster inference. It establishes a new, efficient, and scalable paradigm for graph similarity modeling, particularly suitable for large-scale architectural design comparison and retrieval.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

Floor plans depict building layouts and are often represented as graphs to capture the underlying spatial relationships. Comparison of these graphs is critical for applications like search, clustering, and data visualization. The most successful methods to compare graphs ie, graph matching networks, rely on costly intermediate cross-graph node-level interactions, therefore being slow in inference time. We introduce extbf{LayoutGKN}, a more efficient approach that postpones the cross-graph node-level interactions to the end of the joint embedding architecture. We do so by using a differentiable graph kernel as a distance function on the final learned node-level embeddings. We show that LayoutGKN computes similarity comparably or better than graph matching networks while significantly increasing the speed. href{https://github.com/caspervanengelenburg/LayoutGKN}{Code and data} are open.

Problem

• Efficiently comparing floor plan graphs for similarity
• Reducing slow inference time in graph matching networks
• Postponing cross-graph interactions in joint embedding architecture

Innovation

• Postpones cross-graph node interactions
• Uses differentiable graph kernel
• Improves speed while maintaining accuracy