Deep clustering algorithms combine representation learning and clustering by\njointly optimizing a clustering loss and a non-clustering loss. In such\nmethods, a deep neural network is used for representation learning together\nwith a clustering network. Instead of following this framework to improve\nclustering performance, we propose a simpler approach of optimizing the\nentanglement of the learned latent code representation of an autoencoder. We\ndefine entanglement as how close pairs of points from the same class or\nstructure are, relative to pairs of points from different classes or\nstructures. To measure the entanglement of data points, we use the soft nearest\nneighbor loss, and expand it by introducing an annealing temperature factor.\nUsing our proposed approach, the test clustering accuracy was 96.2% on the\nMNIST dataset, 85.6% on the Fashion-MNIST dataset, and 79.2% on the EMNIST\nBalanced dataset, outperforming our baseline models.\n