Fig. 2: Classification and novelty detection using the reconstruction error.
From: Deep learning for visualization and novelty detection in large X-ray diffraction datasets
The decision boundaries of a KNN classifier are outlined. a Latent space representation of a known phase color-coded by the reconstruction error. b Latent space representation of an unknown phase color-coded by the reconstruction error. The unknown phase shows a distinctly higher reconstruction error (average reconstruction error = 0.09) compared to phases that are recognized by the model (average reconstruction error = 0.017). The classification test score in this idealized case was 99.16%.
