Deep clustering has recently emerged as a promising technique for complex image clustering. Despite the significant progress, previous deep clustering works mostly tend to construct the final clustering by utilizing a single layer of representation, e.g., by performing $K$-means on the last fully-connected layer or by associating some clustering loss to a specific layer. However, few of them have considered the possibilities and potential benefits of jointly leveraging multi-layer representations for enhancing the deep clustering performance. In light of this, this paper presents a Deep Clustering via Ensembles (DeepCluE) approach, which bridges the gap between deep clustering and ensemble clustering by harnessing the power of multiple layers in deep neural networks. Particularly, we utilize a weight-sharing convolutional neural network as the backbone, which is trained with both the instance-level contrastive learning (via an instance projector) and the cluster-level contrastive learning (via a cluster projector) in an unsupervised manner. Thereafter, multiple layers of feature representations are extracted from the trained network, upon which a set of diversified base clusterings can be generated via a highly efficient clusterer. Then, the reliability of the clusters in multiple base clusterings is automatically estimated by exploiting an entropy-based criterion, based on which the multiple base clusterings are further formulated into a weighted-cluster bipartite graph. By partitioning this bipartite graph via transfer cut, the final image clustering result can therefore be obtained. Experimental results on six image datasets confirm the advantages of our DeepCluE approach over the state-of-the-art deep clustering approaches.

本文提出了一种名为 DeepCluE 的方法，它通过利用深度神经网络中的多层特征表示来增强深度聚类性能。该方法利用了无监督学习的实例级对比学习和聚类级对比学习，生成多组基本聚类，通过基于熵准则的可靠性估计，将这些聚类形成加权聚簇的双分图，利用转移割划分该双分图得到最终的图像聚类结果。实验结果表明，相对于先前的深度聚类方法，DeepCluE方法具有明显的优势。

Deep Neural Networks 中多层集合增强的图像聚类 DeepCluE