Mixing Consistent Deep Clustering

Finding well-defined clusters in data represents a fundamental challenge for many data-driven applications, and largely depends on good representation. Drawing literature regarding representation learning, studies suggest that one key characteristic of latent representations is the ability to produce semantically mixed outputs when decoding linear interpolations two representations. We propose Mixing Consistent Deep Clustering (MCDC) method which encourages appear realistic while adding constraint points must look like inputs. By applying this training various clustering (non-)specific autoencoder models we found using proposed systematically changed structure learned model it improved performance tested ACAI, IDEC, VAE MNIST, SVHN, CIFAR-10 datasets. These outcomes have practical implications numerous real-world tasks, as shows can be added existing autoencoders further improve performance.