Semisupervised Deep Embedded Clustering with Adaptive Labels

Figure 3

The visualization of clustering results during training on subset of MNIST, USPS, and REUTERS-10K from top to bottom. Different colours mark different clusters. The clustering accuracy of the corresponding epoch is given in parentheses. It can be seen that the data of the same class become more compact while the data of different classes are further away from each other as the number of epochs increases. This also shows that the learned feature embedding space is more and more suitable for clustering tasks.