In this paper, we propose a novel framework for Deep Clustering and multi-manifold Representation Learning (DCRL) that preserves the geometric structure of data. In the proposed framework, manifold clustering 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, an isometric loss is proposed for preserving intra-manifold structure locally and a ranking loss for inter-manifold structure globally. Experimental results on various datasets show that DCRL leads to performances comparable to current state-of-the-art deep clustering algorithms, yet exhibits superior performance for manifold representation. Our results also demonstrate the importance and effectiveness of the proposed losses in preserving geometric structure in terms of visualization and performance metrics.
翻译:在本文中,我们提议了一个保护数据几何结构的深层集群和多层代表性学习(DCRL)新框架。在拟议框架中,以集群损失为指针,在潜在空间进行多重组合。为了克服以集群为导向的损失可能恶化潜层嵌入层的几何结构的问题,我们提议了一种等计量损失,以维护本地的内层结构,并对全球内层结构进行排序损失。各种数据集的实验结果表明,DCRL导致的性能与当前最先进的深层集群算法相类似,但显示多重代表性的优异性。我们的结果还表明,拟议损失对于维护可视化和性能衡量的几何结构的重要性和有效性。