Font Size: 

Classical clustering methods may suffer from the presence of high dimensional or complex data. In this scenario, deep clustering can be useful to overcome such problems. The main idea is to use a neural network to reduce the input’s complexity and apply a clustering algorithm to the reduced space. Our method consists in combining a neural network with the fuzzy k-means clustering algorithm. In particular, the proposal links the encoder part of an autoencoder neural network to a new layer, in which the membership degree values are calculated, and jointly optimizes the method by minimizing the fuzzy k-means objective function. Furthermore, to avoid the problem of collapsing centers, a penalization term is added. The adequacy of the proposal is evaluated by means of benchmark datasets.