Zhuosheng Lin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Zhuosheng Lin has a total of 39 co-authors. Here are the top co-authors by number of shared publications:
- Qianxue Wang
School of Automation Guangdong University of Technology 510006 Guangzhou Guangdong China
4 shared publications Yue Feng
4 shared publications- Christophe Guyeux
University of Franche-Comte
3 shared publications Jingliang Deng
3 shared publications- Simin Yu
School of Automation, Guangdong University of Technology, Guangzhou 510006, P. R. China
3 shared publications - Yue Feng
School of Engineering, Swinburne University of Technology, Hawthorn, Victoria, Australia
3 shared publications
Latest Publications
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-15
Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-15
Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge
- 25 December 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 335
A plug-and-play steganography method via latent space manipulation for multiple generation modes
- 25 December 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 335
A plug-and-play steganography method via latent space manipulation for multiple generation modes
- 18 June 2025
- Published by Springer Nature in Cognitive Neurodynamics
- Vol. 19 (1) , 1-10
An effective ECG signal classification method based on a minimalistic memristive reservoir computing system
The advantage of reservoir computing (RC) is that only the connection weight between reservoir layer and output layer needs to be trained, and the rest of the connection weights are randomly generated and fixed, which is especially suitable for time series data processing. However, the hardware implementation of high-dimensional random connection of neurons in reservoir layer is still a challenge. The memristors are emerging components with unique nonlinear and memory characteristics. Particularly, memristors allow nonlinear mapping of input time series into high-dimensional feature space, which meets the requirements of the reservoir layer. The reservoir layers of current memristor-based RC systems are based on the dynamics of volatile memristors. In this paper, a non-volatile memristor-based reservoir layer is designed for constructing RC system, in which the voltages across the two memristors are utilized to calculate the reservoir states. In this design, one-dimensional voltage input signal can be easily nonlinearly mapped to a two-dimensional space, which significantly simplifies the complexity of data analysis and enhances the separability of signal features, satisfying the requirement of RC for the high-dimensional feature. The experimental results of the proposed RC system for electrocardiogram (ECG) signal classification task achieve high accuracies of 98.3% and 100% for QRS complexes with and without shift, respectively, which validates the effectiveness of the proposed RC system.
- 18 June 2025
- Published by Springer Nature in Cognitive Neurodynamics
- Vol. 19 (1) , 1-10
An effective ECG signal classification method based on a minimalistic memristive reservoir computing system
The advantage of reservoir computing (RC) is that only the connection weight between reservoir layer and output layer needs to be trained, and the rest of the connection weights are randomly generated and fixed, which is especially suitable for time series data processing. However, the hardware implementation of high-dimensional random connection of neurons in reservoir layer is still a challenge. The memristors are emerging components with unique nonlinear and memory characteristics. Particularly, memristors allow nonlinear mapping of input time series into high-dimensional feature space, which meets the requirements of the reservoir layer. The reservoir layers of current memristor-based RC systems are based on the dynamics of volatile memristors. In this paper, a non-volatile memristor-based reservoir layer is designed for constructing RC system, in which the voltages across the two memristors are utilized to calculate the reservoir states. In this design, one-dimensional voltage input signal can be easily nonlinearly mapped to a two-dimensional space, which significantly simplifies the complexity of data analysis and enhances the separability of signal features, satisfying the requirement of RC for the high-dimensional feature. The experimental results of the proposed RC system for electrocardiogram (ECG) signal classification task achieve high accuracies of 98.3% and 100% for QRS complexes with and without shift, respectively, which validates the effectiveness of the proposed RC system.
- 19 April 2025
- Published by Springer Nature in Nonlinear Dynamics
- Vol. 113 (15) , 20191-20209
An efficient reservoir computing system based on 2D mask processing and dynamic memristor
- 19 April 2025
- Published by Springer Nature in Nonlinear Dynamics
- Vol. 113 (15) , 20191-20209
An efficient reservoir computing system based on 2D mask processing and dynamic memristor
- 6 March 2024
- Published by Springer Nature in Nonlinear Dynamics
- Vol. 112 (8) , 6663-6678
Nonmasking-based reservoir computing with a single dynamic memristor for image recognition
- 6 March 2024
- Published by Springer Nature in Nonlinear Dynamics
- Vol. 112 (8) , 6663-6678
Nonmasking-based reservoir computing with a single dynamic memristor for image recognition
- Open Access
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 11217-11229
An Improved ConvNeXt With Multimodal Transformer for Physiological Signal Classification
With escalating mortality rates associated with cardiovascular disease, the early detection of arrhythmias assumes ever-increasing significance. This study introduces a novel multimodal network that concurrently classifies electrocardiogram (ECG) and wrist pulse signal (WPS). Both ECG and WPS, as human physiological signals, share closely related distributions and characteristics, holding potential to accurately reflect underlying cardiovascular conditions. The proposed ICMT-Net utilizes continuous wavelet transform to partition 5-second ECG and WPS segments into spectrograms. It incorporates an improved ConvNeXt, a multimodal transformer layer, and a fused multi-layer perceptron to extract and fuse multimodal features for ECG classification. Subsequently, the network is adapted to WPS and coronary heart disease classification tasks through transfer learning techniques. In comparison to existing methods, our approach achieves heightened sensitivity in detecting supraventricular and ventricular ectopic segments, while also outperforming established WPS classification methodologies. Importantly, the proposed network adeptly handles multimodal signals and excels in classification accuracy, particularly within the realm of physiological signals.
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 11217-11229
Open AccessAn Improved ConvNeXt With Multimodal Transformer for Physiological Signal Classification
With escalating mortality rates associated with cardiovascular disease, the early detection of arrhythmias assumes ever-increasing significance. This study introduces a novel multimodal network that concurrently classifies electrocardiogram (ECG) and wrist pulse signal (WPS). Both ECG and WPS, as human physiological signals, share closely related distributions and characteristics, holding potential to accurately reflect underlying cardiovascular conditions. The proposed ICMT-Net utilizes continuous wavelet transform to partition 5-second ECG and WPS segments into spectrograms. It incorporates an improved ConvNeXt, a multimodal transformer layer, and a fused multi-layer perceptron to extract and fuse multimodal features for ECG classification. Subsequently, the network is adapted to WPS and coronary heart disease classification tasks through transfer learning techniques. In comparison to existing methods, our approach achieves heightened sensitivity in detecting supraventricular and ventricular ectopic segments, while also outperforming established WPS classification methodologies. Importantly, the proposed network adeptly handles multimodal signals and excels in classification accuracy, particularly within the realm of physiological signals.
- Open Access
A Ferroelectric Memristor-Based Transient Chaotic Neural Network for Solving Combinatorial Optimization Problems
A transient chaotic neural network (TCNN) is particularly useful for solving combinatorial optimization problems, and its hardware implementation based on memristors has attracted great attention recently. Although previously used filamentary memristors could provide the desired nonlinearity for implementing the annealing function of a TCNN, the controllability of filamentary switching still remains relatively poor, thus limiting the performance of a memristor-based TCNN. Here, we propose to use ferroelectric memristor to implement the annealing function of a TCNN. In the ferroelectric memristor, the conductance can be tuned by switching the lattice non-centrosymmetry-induced polarization, which is a nonlinear switching mechanism with high controllability. We first establish a ferroelectric memristor model based on a ferroelectric tunnel junction (FTJ), which exhibits the polarization-modulated tunnel conductance and the nucleation-limited-switching (NLS) behavior. Then, the conductance of the ferroelectric memristor is used as the self-feedback connection weight that can be dynamically adjusted. Based on this, a ferroelectric memristor-based transient chaotic neural network (FM-TCNN) is further constructed and applied to solve the traveling salesman problem (TSP). In 1000 runs for 10-city TSP, the FM-TCNN achieves a shorter average path distance, a 32.8% faster convergence speed, and a 2.44% higher global optimal rate than the TCNN.
Open AccessA Ferroelectric Memristor-Based Transient Chaotic Neural Network for Solving Combinatorial Optimization Problems
A transient chaotic neural network (TCNN) is particularly useful for solving combinatorial optimization problems, and its hardware implementation based on memristors has attracted great attention recently. Although previously used filamentary memristors could provide the desired nonlinearity for implementing the annealing function of a TCNN, the controllability of filamentary switching still remains relatively poor, thus limiting the performance of a memristor-based TCNN. Here, we propose to use ferroelectric memristor to implement the annealing function of a TCNN. In the ferroelectric memristor, the conductance can be tuned by switching the lattice non-centrosymmetry-induced polarization, which is a nonlinear switching mechanism with high controllability. We first establish a ferroelectric memristor model based on a ferroelectric tunnel junction (FTJ), which exhibits the polarization-modulated tunnel conductance and the nucleation-limited-switching (NLS) behavior. Then, the conductance of the ferroelectric memristor is used as the self-feedback connection weight that can be dynamically adjusted. Based on this, a ferroelectric memristor-based transient chaotic neural network (FM-TCNN) is further constructed and applied to solve the traveling salesman problem (TSP). In 1000 runs for 10-city TSP, the FM-TCNN achieves a shorter average path distance, a 32.8% faster convergence speed, and a 2.44% higher global optimal rate than the TCNN.