Learning without relearning physical symmetry: an SO(3)-equivariant convolutional network (CNN) on the sphere
Novel architecture for exact symmetry preservation in spherical CNNs with applications in cosmology and physical sciences
AI Summary
Researchers developed an SO(3)-equivariant CNN for spherical data that preserves physical symmetries, reducing simulation requirements and improving accuracy in cosmology applications.
Excerpt
Many of the most important datasets in the physical sciences live natively on the sphere, and these data often carry no preferred position or orientation. Using methods that do not reflect this symmetry leads to far higher simulation requirements, which can be catastrophic for scientific analyses (forcing reliance on cheap, imperfect models that bias results). We present a novel spherical implementation of a gauge equivariant mesh architecture that is specifically designed for the Hierarchical E