Learning Chern Numbers of Topological Insulators with Gauge Equivariant Neural Networks
This episode details a paper using gauge equivariant neural networks to predict the Chern number, a property that identifies topological insulators. The model successfully handles multiple electron bands—a major improvement over previous methods. Practical enhancements include a novel normalization layer (TrNorm) and specialized loss functions, providing a robust tool for discovering new materials.