Generative Design through Quality-Diversity Data Synthesis and Language Models

Adam Gaier,James Stoddart,Lorenzo Villaggi,Shyam Sudhakaran

DOI: https://doi.org/10.1145/3638529.3654138

2024-05-16

Abstract:Two fundamental challenges face generative models in engineering applications: the acquisition of high-performing, diverse datasets, and the adherence to precise constraints in generated designs. We propose a novel approach combining optimization, constraint satisfaction, and language models to tackle these challenges in architectural design. Our method uses Quality-Diversity (QD) to generate a diverse, high-performing dataset. We then fine-tune a language model with this dataset to generate high-level designs. These designs are then refined into detailed, constraint-compliant layouts using the Wave Function Collapse algorithm. Our system demonstrates reliable adherence to textual guidance, enabling the generation of layouts with targeted architectural and performance features. Crucially, our results indicate that data synthesized through the evolutionary search of QD not only improves overall model performance but is essential for the model's ability to closely adhere to textual guidance. This improvement underscores the pivotal role evolutionary computation can play in creating the datasets key to training generative models for design. Web article at

Neural and Evolutionary Computing,Machine Learning

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

The paper aims to address two fundamental challenges faced by generative models in engineering applications:

1. **Acquiring high-quality, diverse datasets**: During the design process, a large number of high-performance and diverse design samples are needed as training data.
2. **Meeting strict design constraints**: The generated designs need to strictly adhere to specific constraints.

To tackle these challenges, the paper proposes a new method that combines optimization algorithms, constraint satisfaction techniques, and language models to solve problems in architectural design. Specifically, the method utilizes Quality-Diversity (QD) algorithms to generate diverse and high-performance datasets, and uses these datasets to fine-tune language models to generate high-level designs. These designs are then further refined using the Wave Function Collapse (WFC) algorithm to ensure that the final designs meet practical constraints.

The main contributions of the paper include:
- Combining QD with Procedural Content Generation (PCG) techniques to generate customized labeled datasets.
- Using fine-tuned language models to interpret natural language instructions and apply them to actual design cases.
- Applying constraint satisfaction techniques to ensure the validity of designs generated by the language models.

Through this method, the paper demonstrates the reliability of generated designs and their high adherence to textual guidance, proving the importance of evolutionary computation in creating critical training datasets.