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