Deep Clustering and Representation Learning that Preserves Geometric Structures

In this paper, we propose a novel framework for Deep Clustering and Representation Learning (DCRL) that preserves the geometric structure of data. In the proposed DCRL framework, the clustering of data points from different manifolds is done in the latent space guided by a clustering loss. To overcome the problem that clustering-oriented losses may deteriorate the geometric structure of embeddings in the latent space, two structure-oriented losses, namely an isometric loss and a ranking loss, are proposed to preserve intra-manifold structure locally and inter-manifold structure globally. Experimental results on various datasets show that the DCRL framework leads to comparable performances to current state-of-the-art deep clustering algorithms, yet exhibits superior performance for downstream tasks. Our results also demonstrate the importance and effectiveness of the proposed method in preserving geometric structure in terms of visualization and performance metrics.