Fig. 3: Cumulative explained variance versus ranked principal components when using PCA directly on actual microstructure images (ambient space) or on latent representation (latent space).
From: Inferring topological transitions in pattern-forming processes with self-supervised learning
a Explained variance for the spinodal decomposition problem. b Explained variance for the physical vapor deposition problem. Ambient space denotes dimensionality reduction when PCA is performed directly on the high-dimensional microstructure images, while latent space denotes the projection when PCA is performed on the already compact representation of the microstructures obtained from the ResNet model.
