A Study of Deep Fuzzy Clustering Method Based on Maximum Entropy Clustering

Clustering of high-dimensional data is a main research content in the field of data mining and pattern recognition research. Combined with neural network, this paper proposes a deep fuzzy clustering method based on maximum entropy clustering, which using a layer of feed-forward neural network to realize fuzzy clustering and with interpretable for the clustering process. Integrating fuzzy clustering and deep autoencoder model, this method updates network parameters and cluster center through network gradient and realizes deep feature extraction and fuzzy clustering simultaneously. The activation function of the cluster layer as a soft assignment indicator achieves fuzzy cluster matching. The maximum entropy regularization term is used to optimize and adjust the clustering objective, which can enhance the robustness of the model. The method is compared with the classical clustering algorithms on the MNIST and USPS handwritten digit datasets to verify its effectiveness.