Adversarial Incomplete Multi-view Clustering

Abstract

Multi-view clustering aims to leverage information from multiple views to improve clustering. Most previous works assumed that each view has complete data. However, in real-world datasets, it is often the case that a view may contain some missing data, resulting in the incomplete multi-view clustering problem. Previous methods for this problem have at least one of the following drawbacks: (1) employing shallow models, which cannot well handle the dependence and discrepancy among different views; (2) ignoring the hidden information of the missing data; (3) dedicated to the two-view case. To eliminate all these drawbacks, in this work we present an Adversarial Incomplete Multi-view Clustering (AIMC) method. Unlike most existing methods which only learn a new representation with existing views, AIMC seeks the common latent space of multi-view data and performs missing data inference simultaneously. In particular, the element-wise reconstruction and the generative adversarial network (GAN) are integrated to infer the missing data. They aim to capture overall structure and get a deeper semantic understanding respectively. Moreover, an aligned clustering loss is designed to obtain a better clustering structure. Experiments conducted on three datasets show that AIMC performs well and outperforms baseline methods.

DIMC-net: Deep Incomplete Multi-view Clustering Network

“…Clustering layer: The above three parts only focus on obtaining the consensus representation, but cannot guarantee that the obtained representation is suitable for clustering. Inspired by the good properties of 'KL divergence' based clustering [4,25,31,32], we adopt it as the last layer of our DIMC-net, which not only can effectively address the above issue, but also can directly produce the clustering results. Specifically, for the output of the weighted fusion layer, i.e., consensus representation * ∈ × of all samples, if its cluster center is denoted by { } =1 ( ∈ ×1 ), the objective function of 'KL divergence' based clustering layer is expressed as follows [4,31]:…”

“…In recent years, incomplete multi-view clustering (IMC) has received a lot of attention due to the increasing amount of incomplete multi-view data in the real-world applications, such as recommendation system, multimedia analysis, and bioinformatics [5,22,32,39]. For example, in audio and visual based speech analysis, the collected data may only have audio or visual appearance for some speakers.…”

“…To further enhance the IMC performance, some IMC extensions based on deep learning are proposed. For instance, in works [25] and [32], the popular Autoencoder and generative adversarial networks (GANs) are integrated to address the incomplete issue. Different from the other IMC methods, these two methods can recover the missing views and in turn exploit it to promote the consensus representation learning.…”

Federated Multi-View Spectral Clustering

“…Recently, deep learning based methods [9], [19] have been dedicated to incorporating the local invariance within every single view and the consistency across different views. However, all the aforementioned approaches assume the multiview data can be collected on a central server, which will expose their sensitive data to other parties.…”

Semi-Supervised Multi-Modal Learning with Balanced Spectral Decomposition

“…With the rapid growth of multimedia data such as image, text, and audio on the Internet, there are increasing demands on developing various applications to handle this data, such as classification (Guan et al 2015), clustering (Xu et al 2018;Peng et al 2019;Xu et al 2019), and retrieval (Deng et al 2018;Hu et al 2019b). Over the past decades, more and more attention has been attracted by retrieving the interested contents across different modalities, namely cross-modal retrieval (Peng, Qi, and Yuan 2018;Hu et al 2019a).…”