Adversarial Incomplete Multiview Subspace Clustering Networks

Xu, Cai; Liu, Hong‐Min; Guan, Ziyu; Wu, Xunlian; Tan, Jiale; Ling, Beilei

doi:10.1109/tcyb.2021.3062830

Cited by 40 publications

(8 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, Zhao et al [30] developed a novel incomplete multi-view clustering method, which projects all multi-view data to a complete and unified representation with the constrain of intrinsic geometric structure. Wang et al [32] and Xu et al [33] proposed incomplete multi-view clustering methods based on generative adversarial network (GAN) [34], aiming at two-view and multi-view data respectively, i.e., they use existing views to generate missing views. Lin et al [35] introduced the ideal of contrast prediction from self-supervised learning in to incomplete multi-view learning to explore the consensus between views by maximizing mutual information and minimizing conditional entropy.…”

Section: Deep Neural Network Based Methods [29][30][31][32]mentioning

confidence: 99%

Adaptive incomplete multi-view learning via tensor graph completion

Zhang¹,

Chen²

2022

Preprint

View full text Add to dashboard Cite

With the advancement of the data acquisition techniques, multi-view learning has become a hot topic. Some multi-view learning methods assume that the multi-view data is complete, which means that all instances are present, but this too ideal. Certain tensor-based methods for handing incomplete multi-view data have emerged and have achieved better result. However, there are still some problems, such as use of traditional tensor norm which makes the computation high and is not able to handle out-of-sample. To solve these two problems, we proposed a new incomplete multiview learning method. A new tensor norm is defined to implement graph tensor data recover. The recovered graphs are then regularized to a consistent low-dimensional representation of the samples. In addition, adaptive weights are equipped to each view to adjust the importance of different views. Compared with the existing methods, our method nor only explores the consistency among views, but also obtains the low-dimensional representation of the new samples by using the learned projection matrix. An efficient algorithm based on inexact augmented Lagrange multiplier (ALM) method are designed to solve the model and convergence is proved. Experimental results on four datasets show the effectiveness of our method.

show abstract

Section: Deep Neural Network Based Methods [29][30][31][32]mentioning

confidence: 99%

Adaptive incomplete multi-view learning via tensor graph completion

Zhang¹,

Chen²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…With the development of deep learning, a number of recent studies (Zhang et al 2020;Lin et al 2021) employ DNNs to solve incomplete multi-view clustering challenges, showing noticeable improvement on clustering performance. Based on producing missing data or not, most existing deep incomplete multi-view clustering methods could be roughly classified into two categories, i.e., learning a uniform representation without generating missing data (Wen et al 2020b,a;Wang et al 2021b) and generating missing instances using adversarial learning (Wang et al 2021a(Wang et al , 2018Zhang et al 2020;Xu et al 2021). The first class of methods reduces the negative influence of missing views by modifying the multi-view fusion representation learning model.…”

Section: Related Workmentioning

confidence: 99%

Incomplete Multi-view Clustering via Cross-view Relation Transfer

Wang¹,

Chang²,

Fu³

et al. 2021

Preprint

View full text Add to dashboard Cite

In this paper, we consider the problem of multi-view clustering on incomplete views. Compared with complete multiview clustering, the view-missing problem increases the difficulty of learning common representations from different views. To address the challenge, we propose a novel incomplete multi-view clustering framework, which incorporates cross-view relation transfer and multi-view fusion learning. Specifically, based on the consistency existing in multi-view data, we devise a cross-view relation transfer-based completion module, which transfers known similar inter-instance relationships to the missing view and recovers the missing data via graph networks based on the transferred relationship graph. Then the view-specific encoders are designed to extract the recovered multi-view data, and an attention-based fusion layer is introduced to obtain the common representation. Moreover, to reduce the impact of the error caused by the inconsistency between views and obtain a better clustering structure, a joint clustering layer is introduced to optimize recovery and clustering simultaneously. Extensive experiments conducted on several real datasets demonstrate the effectiveness of the proposed method.

show abstract

“…CLUSTERING RESULTS AND STANDARD DEVIATIONS OF THE DIFFERENT METHODS WITH VARIOUS MISSING RATIOS ON SIX MULTIVIEWDATASETS. Agg 91.97±0.00 90.22±0.00 89.64±0.00 81.75±0.00 77.96±0.00 74.75±0.00 72.15±0.00 56.15±0.00 91.99±0.00 90.2±0.00 89.61±0.00 81.48±0.00 DAIMC 81.02±0.00 75.18±0.00 73.86±0.00 68.46±0.00 65.3±0.00 63.01±0.00 48.66±0.0042.98±0.00 81.71±0.00 76.57±0.00 75.35±0.00 69.92±0.00 SRSC 91.11±0.05 91.1±0.00 89.2±0.00 82.63±0.00 75.71±0.10 75.46±0.00 70.81±0.00 56.73±0.00 91.11±0.05 91.07±0.00 89.19±0.00 82.74±0.00 GIMC 83.94±0.00 82.04±0.00 82.92±0.00 77.23±0.00 67.7±0.00 63.27±0.00 64.89±0.00 58.56±0.00 83.53±0.00 81.86±0.00 82.71±0.00 77.78±0.00 EE-IMVC 70.02±4.74 70.89±4.96 68.89±4.37 58.32±2.91 54.07±2.71 53.97±2.91 49.97±2.60 39.54±1.66 72.7±2.69 72.87±2.88 70.61±2.68 64.14±0.78 ANIMC 72.85±2.99 50.5±4.22 39.74±3.49 34.41±2.83 51.42±3.21 28.01±3.54 15.83±2.23 11.1±2.76 73.15±2.87 55.14±3.07 45.91±1.62 37.IMVC 59.2±1.92 58.63±1.29 52.67±1.64 48.63±1.06 57.33±1.20 55.12±0.91 49.89±0.84 43.29±0.62 62.64±1.93 61.19±1.18 55.03±1.46 51.12±1.16 ANIMC 45.97±1.54 37.93±1.24 34.32±1.41 31.94±2.23 47.04±1.81 36.62±1.25 32.92±0.57 29.32±1.37 49.77±2.22 41.13±1.34 37.02±1.71 34.78±2.46 ASR 68.27±0.19 64.59±0.17 62.07±0.26 52.17±0.22 65.63±0.20 61.94±0.16 58.8±0.23 48.1±0.32 71.16±0.17 68.17±0.2 64.78±0.26 55.56±0.19 ProteinFold SSC Agg 25.99±1.09 24.42±0.78 24.65±0.64 19.27±0.79 34.28±0.76 32.35±0.70 32.52±0.72 26.36±0.59 27.82±1.32 26.3±0.85 25.96±0.4 20.95±0.87 DAIMC 28.26±1.43 27.51±1.88 27.28±1.5 24.58±0.92 39.72±0.47 38.94±1.7 37.16±0.97 33.38±0.65 31.47±1.12 30.45±1.86 30.33±1.58 27.27±0.866 SRSC 37.06±1.44 34.04±1.54 34.58±1.43 31.38±0.70 45.97±0.79 44.71±0.69 43.12±0.75 39.13±0.80 39.75±1.30 37.5±1.33 37.01±1.44 33.52±0.83 GIMC 27.52±0.00 24.5±0.00 23.78±0.00 21.18±0.00 37.51±0.00 33.6±0.00 30.88±0.00 29.62±0.00 30.48±0.00 27.4±0.00 27.26±0.00 23.87±0.00 EE-IMVC 33.07±1.9 34.4±2.20 32.13±2.22 29.71±1.82 43.25±0.78 43.63±1.13 41.32±1.56 37.9±0.71 35.68±1.87 36.69±1.97 34.53±2.37 31.54±1.74 ANIMC 16.76±0.79 14.77±0.57 13.62±0.55 12.85±0.43 24.9±1.35 20.54±0.61 19.37±1.02 17.67±0.5 18.6±0.89 16.3±0.61 14.78±0.65 13.49±0.5 ASR 41.43±0.57 41.25±1.19 36.73±1.05 34.64±1.16 50.26±0.57 50.03±0.48 46.69±0.98 42.82±0.81 43.33±0.56 43.59±1.04 39.89±1.09 37.51±1.17 100leaves SSC Agg 80.66±1.19 70.69±1.08 54.15±1.62 33.33±0.72 90.56±0.42 84.71±0.36 74.15±0.56 60.69±0.35 82.02±0.98 73.18±0.87 57.77±1.27 37.8±0.78 DAIMC 77.85±2.02 60.21±1.66 45.69±2.13 31.91±1.16 90.95±1.08 80.7±1.02 69.67±1.14 58.17±0.55 80.79±1.87 64.76±1.64 50.6±1.82 36.96±1.11 SRSC 81.02±1.16 70.71±1.14 51.16±1.48 33.88±0.85 91.94±0.68 84.82±0.41 72.66±0.59 60.93±0.56 83.57±1.14 74.58±1.00 56.33±0.87 38.98±0.78 GIMC 79.63±0.00 66.25±0.00 46.31±0.00 24.44±0.00 91.79±0.00 84.92±0.00 71.43±0.00 53.54±0.00 82.64±0.00 69.55±0.00 49.92±0.00 27.41±0.00 EE-IMVC 74.76±1.03 64.05±1.65 43.68±1.29 31.12±1.05 88.25±0.67 80.6±0.71 67.9±0.62 59.47±0.57 77.03±0.95 66.92±1.56 47.98±1.21 35.45±0.98 ANIMC 65.425±1.79 26.35±0.86 21.9±0.75 17.9±0.80 83.41±0.81 54.46±0.570 48.6±0.71 43.02±0.51 08.48±1.42 30.57±0.73 26.33±0.83 23.35±0.75 ASR 86.41±1.09 78.04±0.73 56.31±0.81 40.52±0.32 95.14±0.49 89.38±0.42 74.34±0.3 62.65±0.44 88.83±0.92 80.99±0.64 60.94±0.69 46.84±0.47 Agg 25.98±0.51 23.51±0.51 20.63±0.49 18.25±0.37 47.28±0.12 44.63±0.20 41.26±0.19 34.22±0.17 30.76±0.48 28.3±0.67 24.53±0.54 20.93±0.60 DAIMC 26.1±0.68 25.46±0.87 21.15±1.36 18.7±1.05 51.55±0.29 50.57±0.34 43.17±0.7 36.37±0.88 32.85±0.82 31.71±1.1 27.53±1.55 23.91±1.04 SRSC 23.73±0.53 20±0.4 21.9 ±0.53 18.27±0.55 48.47±0.38 42.92±0.21 42.26±0.21 37.01±0.41 29.99±0.57 26.39±0.61 27.65±0.52 23.01±0.69 GIMC 18.11±0.00 16.09±0.00 15.02±0.00 13.14±0.00 39.05±0.00 36.02±0.00 35.1±0.00 30.52±0.00 23±0.00 19.61±0.00 18.24±0.00 16.18±0.00 EE-IMVC 27.12±0.88 26.57±0.57 21.36±0.92 18.9±0.55 51.67±0.33 50.16±0.34 43.56±0.19 36.76±0.34 33.21±1.11 31.61±0.61 27.41±0.87 23.4±0.57 ANIMC 15.21±0.57 9.73±0.09 9.45±0.51 9.46±0.23 35.27±0.35 12.36±0.3...…”

mentioning

confidence: 99%

“…Recently, some IMVC methods have been proposed to alleviate the above problem. From the perspective of clustering techniques, existing IMVC methods are mainly divided into five categories, subspace learning-based methods [41], [43], nonnegative matrix factorization (NMF)-based methods [15], [16], graph learning-based methods [20], [39], multiple kernelbased methods [24], [23] and deep learning-based methods [38], [42], [21], [44], [46]. Wen et al present an extension of a low-rank representation (LRR) model that incorporates feature space-based missing-view inferring and manifold space-based similarity graph learning [41].…”

mentioning

confidence: 99%

“…matrix by imputing each incomplete base matrix generated by the incomplete views, where the process of completing the missing features is replaced with imputing the incomplete kernel matrices [23]. Xu et al integrated elementwise reconstruction into a generative adversarial network to infer missing data in multiview data [42]. Although these methods can obtain promising results, there are still some limitations and drawbacks in IMVC.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Augmented Sparse Representation for Incomplete Multiview Clustering

Chen

Yang

Peng

et al. 2024

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

Incomplete multiview data are collected from multiple sources or characterized by multiple modalities, where the features of some samples or some views may be missing. Incomplete multiview clustering aims to partition the data into different groups by taking full advantage of the complementary information from multiple incomplete views. Most existing methods based on matrix factorization or subspace learning attempt to recover the missing views or perform imputation of the missing features to improve clustering performance. However, this problem is intractable due to a lack of prior knowledge, e.g., label information or data distribution, especially when the missing views or features are completely damaged. In this paper, we proposed an augmented sparse representation (ASR) method for incomplete multiview clustering. We first introduce a discriminative sparse representation learning (DSRL) model, which learns the sparse representations of multiple views as applied to measure the similarity of the existing features. The DSRL model explores complementary and consistent information by integrating the sparse regularization item and a consensus regularization item, respectively. Simultaneously, it learns a discriminative dictionary from the original samples. The sparsity constrained optimization problem in the DSRL model can be efficiently solved by the alternating direction method of multipliers. Then, we present a similarity fusion scheme, namely, a sparsity augmented fusion of sparse representations, to obtain a sparsity augmented similarity matrix across different views for spectral clustering. Experimental results on several datasets demonstrate the effectiveness of the proposed ASR method for incomplete multiview clustering.

show abstract

Adversarial Incomplete Multiview Subspace Clustering Networks

Cited by 40 publications

References 51 publications

Adaptive incomplete multi-view learning via tensor graph completion

Adaptive incomplete multi-view learning via tensor graph completion

Incomplete Multi-view Clustering via Cross-view Relation Transfer

Augmented Sparse Representation for Incomplete Multiview Clustering

Contact Info

Product

Resources

About