representation learning

Geometry Matters: Understanding Neural Networks Through Manifolds

Researchers use Riemannian geometry to analyze the intrinsic structure of neural network representations, revealing patterns hidden by traditional similarity metrics.