Yu Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yu Zhang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Jie Ding
Department of Electronic Engineering, School of Information Science and Engineering, Adaptive Networks and Control Laboratory, Fudan University, Shanghai, China
4 shared publications - Xiang Li
Research Institute of Intelligent Complex Systems, Fudan University, Shanghai, China
3 shared publications - Jia Li
Department of Electronic Engineering, School of Information Science and Engineering, Adaptive Networks and Control Laboratory, Fudan University, Shanghai, China
2 shared publications - Binbin Zou
Shanghai Acoustic Laboratory, Chinese Academy of Sciences, Shanghai, China
1 shared publication - Changyun Wen
Nanyang Technological University
1 shared publication - Pan Wang
Department of Electronic Engineering, School of Information Science and Engineering, Adaptive Networks and Control Laboratory, Fudan University, Shanghai, China
1 shared publication
Latest Publications
- 29 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cybernetics
- Vol. PP (21682267) , 1-11
Diffusion Graph Transformer for Learning Controllability Robustness in Large-Scale Networks
Network controllability robustness refers to the ability of networks to maintain controllability in the face of various attacks, which is typically determined by recording a series of controllability values under sustained attacks in simulations. However, conducting such simulation experiments is often extremely time-consuming. Particularly, for large-scale networks, such simulation experiments are even infeasible. Therefore, in order to replace the traditional simulations, we propose a method named diffusion graph transformer (DGT) in this work for learning controllability robustness of complex networks under batch attacks by leveraging the descriptive capability of the graph attention mechanism in node feature representation and the effectiveness of the diffusion principle in feature propagation. DGT first generates embeddings based on node degree attributes, then adaptively propagates the transformed features through a specially designed diffusion strategy, followed by a fully connected layer for the final prediction. Extensive experimental results demonstrate that: 1) DGT exhibits excellent accuracy and significant speed advantages in the task of controllability robustness learning of complex networks; 2) compared to current related research on complex network robustness learning tasks, DGT is the first to address the controllability robustness learning task under batch attacks on large-scale networks, and it demonstrates strong generalization performance in terms of network scale, being applicable to networks with hundreds of thousands of nodes; and 3) the DGT model excels in transferability in terms of attack batch sizes, offering flexibility in predicting network controllability robustness toward various attack batches.
- 29 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cybernetics
- Vol. PP (21682267) , 1-11
Diffusion Graph Transformer for Learning Controllability Robustness in Large-Scale Networks
Network controllability robustness refers to the ability of networks to maintain controllability in the face of various attacks, which is typically determined by recording a series of controllability values under sustained attacks in simulations. However, conducting such simulation experiments is often extremely time-consuming. Particularly, for large-scale networks, such simulation experiments are even infeasible. Therefore, in order to replace the traditional simulations, we propose a method named diffusion graph transformer (DGT) in this work for learning controllability robustness of complex networks under batch attacks by leveraging the descriptive capability of the graph attention mechanism in node feature representation and the effectiveness of the diffusion principle in feature propagation. DGT first generates embeddings based on node degree attributes, then adaptively propagates the transformed features through a specially designed diffusion strategy, followed by a fully connected layer for the final prediction. Extensive experimental results demonstrate that: 1) DGT exhibits excellent accuracy and significant speed advantages in the task of controllability robustness learning of complex networks; 2) compared to current related research on complex network robustness learning tasks, DGT is the first to address the controllability robustness learning task under batch attacks on large-scale networks, and it demonstrates strong generalization performance in terms of network scale, being applicable to networks with hundreds of thousands of nodes; and 3) the DGT model excels in transferability in terms of attack batch sizes, offering flexibility in predicting network controllability robustness toward various attack batches.
- 17 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
Multi-Scale Graph Channel Attention Detectors for Sonar Images in Smart Ocean
Object detection in sonar images is challenging since the performance based on generic object detectors is generally not well on sonar datasets and the prior for the scale distribution of the objects is usually ignored. To overcome these issues, in this article, multi-scale graph channel attention (MGCA) detectors are proposed, in which Multi-Scale Box Copy-Paste is designed for data augmentation and Graph Transformer Channel Attention (GTCA) based on ConvNeXt is introduced to learn powerful visual representations for sonar images. GTCA contains three basic modules: graph structure transformation module for feature maps, graph convolution module for aggregating and updating information and feed forward network module for feature transformation. With Multi-Scale Box Copy-Paste to balance scale distribution and channel attention network in a graph form to recalibrate feature maps, MGCA detectors can effectively detect multi-scale objects in sonar images through transfer learning. The MGCA detector with Cascade R-CNN detection head achieves detection accuracies of 95.6 mAP and 72.7 mmAP on sonar common target detection dataset (SCTD), outperforming the previous best detector with increase of 2.8 mAP and 11.8 mmAP. Extensive experimental results and ablation studies on two sonar datasets demonstrate the superiority of the proposed method compared with state-of-the-art methods.
- 17 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
Multi-Scale Graph Channel Attention Detectors for Sonar Images in Smart Ocean
Object detection in sonar images is challenging since the performance based on generic object detectors is generally not well on sonar datasets and the prior for the scale distribution of the objects is usually ignored. To overcome these issues, in this article, multi-scale graph channel attention (MGCA) detectors are proposed, in which Multi-Scale Box Copy-Paste is designed for data augmentation and Graph Transformer Channel Attention (GTCA) based on ConvNeXt is introduced to learn powerful visual representations for sonar images. GTCA contains three basic modules: graph structure transformation module for feature maps, graph convolution module for aggregating and updating information and feed forward network module for feature transformation. With Multi-Scale Box Copy-Paste to balance scale distribution and channel attention network in a graph form to recalibrate feature maps, MGCA detectors can effectively detect multi-scale objects in sonar images through transfer learning. The MGCA detector with Cascade R-CNN detection head achieves detection accuracies of 95.6 mAP and 72.7 mmAP on sonar common target detection dataset (SCTD), outperforming the previous best detector with increase of 2.8 mAP and 11.8 mmAP. Extensive experimental results and ablation studies on two sonar datasets demonstrate the superiority of the proposed method compared with state-of-the-art methods.
- 22 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 11 (5) , 4045-4058
Network Controllability Robustness Learning via Spatial Graph Neural Networks
Controllability of complex networks refers to the ability of a networked system to steer from any initial state to any desired state in a finite time under feasible control inputs. Its robustness reflects the ability of a network to maintain controllability under various attacks, which records a series of controllability values under continuous attacks and can be directly determined by attack simulations. However, such simulation experiments are very time-consuming. It is necessary to design a model to replace simulation experiments to learn network controllability robustness with high precision and high speed, which is the objective of Network Controllability Robustness Learning . In this paper, given the superiority of spatial graph neural networks in generating network embedding, a novel controllability robustness learning approach via spatial graph neural networks (CRL-SGNN) is proposed by taking into account both generalization and high precision. In this scheme, controllability robustness related complex network representations that contain network topology information and degree-based node attribute information are first generated through the introduced graph convolutional layer and then sent to the prediction module consisting of multiple branches. A large number of experiments show that: 1) CRL-SGNN can overcome the issue of different distributions between training and test sets with satisfactory performance, implying strong generalization ability of CRL-SGNN; 2) Compared to the cutting-edge methods, CRL-SGNN can obtain superior results in less time; and 3) A highlight which cannot be ignored is that the embedding generation module of CRL-SGNN can serve as a transferable feature learning module to deal with complex networks of any size with low prediction error and high efficiency.
- 22 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 11 (5) , 4045-4058
Network Controllability Robustness Learning via Spatial Graph Neural Networks
Controllability of complex networks refers to the ability of a networked system to steer from any initial state to any desired state in a finite time under feasible control inputs. Its robustness reflects the ability of a network to maintain controllability under various attacks, which records a series of controllability values under continuous attacks and can be directly determined by attack simulations. However, such simulation experiments are very time-consuming. It is necessary to design a model to replace simulation experiments to learn network controllability robustness with high precision and high speed, which is the objective of Network Controllability Robustness Learning . In this paper, given the superiority of spatial graph neural networks in generating network embedding, a novel controllability robustness learning approach via spatial graph neural networks (CRL-SGNN) is proposed by taking into account both generalization and high precision. In this scheme, controllability robustness related complex network representations that contain network topology information and degree-based node attribute information are first generated through the introduced graph convolutional layer and then sent to the prediction module consisting of multiple branches. A large number of experiments show that: 1) CRL-SGNN can overcome the issue of different distributions between training and test sets with satisfactory performance, implying strong generalization ability of CRL-SGNN; 2) Compared to the cutting-edge methods, CRL-SGNN can obtain superior results in less time; and 3) A highlight which cannot be ignored is that the embedding generation module of CRL-SGNN can serve as a transferable feature learning module to deal with complex networks of any size with low prediction error and high efficiency.
- 30 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (5) , 1992-2005
A Graph Transformer-Driven Approach for Network Robustness Learning
Learning and analysis of network robustness, including controllability robustness and connectivity robustness, is critical for various networked systems against attacks. Traditionally, network robustness is determined by attack simulations, which is very time-consuming and even incapable for large-scale networks. Network Robustness Learning, which is dedicated to learning network robustness with high precision and high speed, provides a powerful tool to analyze network robustness by replacing simulations. In this paper, a novel versatile and unified robustness learning approach based on the customized graph transformer (NRL-GT) is proposed, which consists of a particularly designed backbone and three branches to accomplish the tasks of robustness curve learning, overall robustness learning and synthetic network classification simultaneously. Numerous experiments show that: 1) NRL-GT is a unified learning framework for controllability robustness and connectivity robustness, demonstrating a strong generalization ability to ensure high precision when training and test sets are distributed differently; 2) It is theoretically and experimentally demonstrated the proposed graph transformer layer is capable of outperforming classical graph neural networks. Compared to the cutting-edge methods in network robustness learning, NRL-GT can simultaneously perform network robustness learning from multiple aspects and obtains superior results in less time on both synthetic and real-world networks, especially circuit networks and power networks; 3) It is worth mentioning that with the transferability of the proposed backbone, NRL-GT is able to deal with complex networks of different sizes for different tasks efficiently and effectively.
- 30 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (5) , 1992-2005
A Graph Transformer-Driven Approach for Network Robustness Learning
Learning and analysis of network robustness, including controllability robustness and connectivity robustness, is critical for various networked systems against attacks. Traditionally, network robustness is determined by attack simulations, which is very time-consuming and even incapable for large-scale networks. Network Robustness Learning, which is dedicated to learning network robustness with high precision and high speed, provides a powerful tool to analyze network robustness by replacing simulations. In this paper, a novel versatile and unified robustness learning approach based on the customized graph transformer (NRL-GT) is proposed, which consists of a particularly designed backbone and three branches to accomplish the tasks of robustness curve learning, overall robustness learning and synthetic network classification simultaneously. Numerous experiments show that: 1) NRL-GT is a unified learning framework for controllability robustness and connectivity robustness, demonstrating a strong generalization ability to ensure high precision when training and test sets are distributed differently; 2) It is theoretically and experimentally demonstrated the proposed graph transformer layer is capable of outperforming classical graph neural networks. Compared to the cutting-edge methods in network robustness learning, NRL-GT can simultaneously perform network robustness learning from multiple aspects and obtains superior results in less time on both synthetic and real-world networks, especially circuit networks and power networks; 3) It is worth mentioning that with the transferability of the proposed backbone, NRL-GT is able to deal with complex networks of different sizes for different tasks efficiently and effectively.