Wenjie Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenjie Zhu has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Wei Qi Yan
Auckland University of Technology, Auckland, New Zealand
7 shared publications Bo Peng
The University of Queensland, Brisbane, Australia
4 shared publications- Huijuan Lu
Main Campus, China Jiliang University, Hangzhou 310018, China
3 shared publications Wei Qi Yan
3 shared publications- Bo Peng
The University of Queensland, Saint Lucia, Brisbane, QLD, Australia
2 shared publications Chunchun Chen
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14616-14625
MoM: Multi-to-Multi Identification of Decoupled Multifaceted Clustering with Dynamic Representation Fusion
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14616-14625
Open AccessMoM: Multi-to-Multi Identification of Decoupled Multifaceted Clustering with Dynamic Representation Fusion
- Open Access
- 1 May 2026
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 3242 (1)
Deep Multi-Modal Non-Redundant Subspace Clustering By Contrastive Affinity Learning
Most approaches to multi-modal subspace clustering aim to identify modality-specific subspaces and subsequently converge them into a consensus subspace. However, real-world data often presents diverse clustering solutions that focus on alternative underlying structures. In this paper, we propose a deep Multi-Modal Non-Redundant Subspace Clustering (MMNRSC) framework, which considers both cross-modality contrastive subspace preservation and the non-redundancy between the consensus subspace and the given clustering solutions by learning contrastive affinities. Specifically, modality-specific networks are designed to extract intermediate features from all modalities, which are more robust than the raw features. To leverage complementary information from all modalities, we introduce a cross-modality contrastive loss based on modality-specific coefficients. This loss function pulls positive pairs together while pushing negative pairs apart, enhancing the modality coupling effect and reinforcing the subspace-preserving property. Additionally, the proposed MMNRSC framework eliminates the memory-intensive self-expressive layer. Experiments conducted on four publicly available datasets demonstrate the effectiveness of the proposed MMNRSC approach in comparison to competitive non-redundant clustering methods.
- 1 May 2026
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 3242 (1)
Open AccessDeep Multi-Modal Non-Redundant Subspace Clustering By Contrastive Affinity Learning
Most approaches to multi-modal subspace clustering aim to identify modality-specific subspaces and subsequently converge them into a consensus subspace. However, real-world data often presents diverse clustering solutions that focus on alternative underlying structures. In this paper, we propose a deep Multi-Modal Non-Redundant Subspace Clustering (MMNRSC) framework, which considers both cross-modality contrastive subspace preservation and the non-redundancy between the consensus subspace and the given clustering solutions by learning contrastive affinities. Specifically, modality-specific networks are designed to extract intermediate features from all modalities, which are more robust than the raw features. To leverage complementary information from all modalities, we introduce a cross-modality contrastive loss based on modality-specific coefficients. This loss function pulls positive pairs together while pushing negative pairs apart, enhancing the modality coupling effect and reinforcing the subspace-preserving property. Additionally, the proposed MMNRSC framework eliminates the memory-intensive self-expressive layer. Experiments conducted on four publicly available datasets demonstrate the effectiveness of the proposed MMNRSC approach in comparison to competitive non-redundant clustering methods.
- 19 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Community detection attack based on adversarial matrix factorization with incoherent embeddings
- 19 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Community detection attack based on adversarial matrix factorization with incoherent embeddings
- 14 March 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (6)
Multi-view co-clustering with dynamic feature-level clustering candidates
- 14 March 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (6)
Multi-view co-clustering with dynamic feature-level clustering candidates
- 10 February 2026
- Published by Elsevier in Information Sciences
- Vol. 740
MrNEAD: Community detection attacks in social networks using modularity-regularized network embedding via adversarial decomposition
- 10 February 2026
- Published by Elsevier in Information Sciences
- Vol. 740
MrNEAD: Community detection attacks in social networks using modularity-regularized network embedding via adversarial decomposition
- 1 November 2025
- Published by Elsevier in Information Sciences
- Vol. 729
Balanced clustering-regularized bilateral spectral learning for unsupervised feature selection in internet of things environments
- 1 November 2025
- Published by Elsevier in Information Sciences
- Vol. 729
Balanced clustering-regularized bilateral spectral learning for unsupervised feature selection in internet of things environments
- 24 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (3) , 3624-3637
Deep Inductive and Scalable Subspace Clustering via Nonlocal Contrastive Self-Distillation
Deep subspace clustering has demonstrated remarkable results by leveraging the nonlinear subspace assumption. However, it often encounters challenges in terms of computational cost and memory footprint in dealing with large-scale data due to its traditional single-batch training strategy. To address this issue, this paper proposes a deep subspace clustering framework that is regularized by nonlocal contrastive self-distillation, enabling a Deep Inductive and Scalable Subspace Clustering (DISSC) algorithm. In particular, our framework incorporates two subspace learning modules, namely subspace learning based on self-expression model and inductive subspace clustering. These modules generate affinities from different perspectives by extracting intermediate features from two augmentations of the input data using a weight-sharing neural network. By integrating the concept of self-distillation, our framework effectively exploits the clustering-friendly knowledge contained in these two affinities through a novel nonlocal contrastive prediction task, employing an empirical yet effective threshold. This allows the framework to facilitate complementary knowledge mining and scalability without compromising clustering performance. With an alternate branch that bypasses the self-expression computation, our framework can infer subspace membership of the out-of-sample data through the predicted soft labels, eliminating the need for ad-hoc postprocessing. In addition, the self-expression matrix computed using mini-batch data benefits from the distilled knowledge obtained from the inductive subspace clustering module, enabling our framework to scale to data of arbitrary size. Experiments conducted on large-scale MNIST, Fashion-MINST, STL-10, CIFAR-10 and Stanford Online Products datasets validate the superiority of the proposed DISSC algorithm over state-of-the-art subspace clustering methods.
- 24 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (3) , 3624-3637
Deep Inductive and Scalable Subspace Clustering via Nonlocal Contrastive Self-Distillation
Deep subspace clustering has demonstrated remarkable results by leveraging the nonlinear subspace assumption. However, it often encounters challenges in terms of computational cost and memory footprint in dealing with large-scale data due to its traditional single-batch training strategy. To address this issue, this paper proposes a deep subspace clustering framework that is regularized by nonlocal contrastive self-distillation, enabling a Deep Inductive and Scalable Subspace Clustering (DISSC) algorithm. In particular, our framework incorporates two subspace learning modules, namely subspace learning based on self-expression model and inductive subspace clustering. These modules generate affinities from different perspectives by extracting intermediate features from two augmentations of the input data using a weight-sharing neural network. By integrating the concept of self-distillation, our framework effectively exploits the clustering-friendly knowledge contained in these two affinities through a novel nonlocal contrastive prediction task, employing an empirical yet effective threshold. This allows the framework to facilitate complementary knowledge mining and scalability without compromising clustering performance. With an alternate branch that bypasses the self-expression computation, our framework can infer subspace membership of the out-of-sample data through the predicted soft labels, eliminating the need for ad-hoc postprocessing. In addition, the self-expression matrix computed using mini-batch data benefits from the distilled knowledge obtained from the inductive subspace clustering module, enabling our framework to scale to data of arbitrary size. Experiments conducted on large-scale MNIST, Fashion-MINST, STL-10, CIFAR-10 and Stanford Online Products datasets validate the superiority of the proposed DISSC algorithm over state-of-the-art subspace clustering methods.
- 17 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 2236-2250
Centrality-Aware Collaborative Network Embedding for Overlapping Community Detection
Community detection is crucial for uncovering the intricate structures and dynamics within various networks, ranging from social interactions to biological systems. Despite significant advancements in community detection approaches, the task of learning network embeddings for complex networks characterized by overlapping communities presents considerable challenges. Specifically, nodes with higher centrality tend to contribute to increased overlaps across multiple communities, complicating the network embedding process. In this paper, we propose a Centrality-Aware Collaborative Learning (CACL) framework that leverages higher-order information through collaborative learning, incorporating a centrality penalty for overlapping community detection. The CACL framework integrates symmetric non-negative matrix factorization and kernel regression models, effectively addressing the limitations associated with traditional single-model techniques. Additionally, it emphasizes the contributions of nodes with higher centrality in the collaborative paradigm. By employing both first-order and second-order information, CACL preserves the intrinsic structure of the network, allowing for a more comprehensive representation of community relationships. To optimize the learning process of CACL, we implement a coordinate descent optimization scheme that tailors network embeddings specifically for overlapping community detection, thereby avoiding ad-hoc processing methods. Extensive experiments on ten LFR benchmark networks and six real-world networks demonstrate that CACL outperforms competing methods in accurately identifying overlapping communities, highlighting its efficacy in centrality-aware collaborative learning.
- 17 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 2236-2250
Centrality-Aware Collaborative Network Embedding for Overlapping Community Detection
Community detection is crucial for uncovering the intricate structures and dynamics within various networks, ranging from social interactions to biological systems. Despite significant advancements in community detection approaches, the task of learning network embeddings for complex networks characterized by overlapping communities presents considerable challenges. Specifically, nodes with higher centrality tend to contribute to increased overlaps across multiple communities, complicating the network embedding process. In this paper, we propose a Centrality-Aware Collaborative Learning (CACL) framework that leverages higher-order information through collaborative learning, incorporating a centrality penalty for overlapping community detection. The CACL framework integrates symmetric non-negative matrix factorization and kernel regression models, effectively addressing the limitations associated with traditional single-model techniques. Additionally, it emphasizes the contributions of nodes with higher centrality in the collaborative paradigm. By employing both first-order and second-order information, CACL preserves the intrinsic structure of the network, allowing for a more comprehensive representation of community relationships. To optimize the learning process of CACL, we implement a coordinate descent optimization scheme that tailors network embeddings specifically for overlapping community detection, thereby avoiding ad-hoc processing methods. Extensive experiments on ten LFR benchmark networks and six real-world networks demonstrate that CACL outperforms competing methods in accurately identifying overlapping communities, highlighting its efficacy in centrality-aware collaborative learning.
- 26 July 2025
- Published by Springer Nature
- p. 412-423
Modularity-Guided Adversarial Matrix Factorization for Community Detection Attacks
- 26 July 2025
- Published by Springer Nature
- p. 412-423
Modularity-Guided Adversarial Matrix Factorization for Community Detection Attacks
- 26 July 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 296
Inductive multiple clustering based on weakly-supervised salient representation learning
- 26 July 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 296
Inductive multiple clustering based on weakly-supervised salient representation learning