Use UMAP to reduce high-dimensional embeddings (e.g., 784-dim FashionMNIST) into 2D/3D to visualize and identify semantic clusters such as trousers, dresses, and footwear.
