Concept Activation Vectors for Generating User-Defined 3D Shapes
This addresses the interpretability challenge in 3D geometric deep learning for CAD users, though it is incremental as it adapts existing CAV methods to a new domain.
The paper tackles the problem of expressing high-level design concepts in parametric CAD by using deep learning to encode 3D shapes into a latent representation, enabling modification of designs based on user-defined concepts with statistical validation.
We explore the interpretability of 3D geometric deep learning models in the context of Computer-Aided Design (CAD). The field of parametric CAD can be limited by the difficulty of expressing high-level design concepts in terms of a few numeric parameters. In this paper, we use a deep learning architectures to encode high dimensional 3D shapes into a vectorized latent representation that can be used to describe arbitrary concepts. Specifically, we train a simple auto-encoder to parameterize a dataset of complex shapes. To understand the latent encoded space, we use the idea of Concept Activation Vectors (CAV) to reinterpret the latent space in terms of user-defined concepts. This allows modification of a reference design to exhibit more or fewer characteristics of a chosen concept or group of concepts. We also test the statistical significance of the identified concepts and determine the sensitivity of a physical quantity of interest across the dataset.