“…Deep clustering is usually performed on the observations (input space) [25], but-may also be applied on the latent (intermediary) representation space [16,7,41,6,18,42,26]. The options for the clustering loss are numerous: k-means loss [40], cluster assignment hardening [39], locality-preserving loss [16], cluster classification loss [15] or agglomerative clustering loss [41] to cite a few.…”