Xiangyu Zeng
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
Research Subjects
Xiangyu Zeng has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Chaoying Yang
School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China
1 shared publication Chufan Chen
Center for Correlated Matter and Department of Physics, Zhejiang University, Hangzhou, Zhejiang, CHINA
1 shared publicationDeWang Chen
Beijing Jiaotong University
1 shared publication- Ge Ye
Center for Correlated Matter, School of Physics, Zhejiang University, Hangzhou 310058, China
1 shared publication GuiWen Jia
Beijing Jiaotong University
1 shared publicationHaoze Kuang
Department of Information Science and Electronic Engineering, Zhejiang University, Hangzhou, Zhejiang, China
1 shared publication
Latest Publications
- 7 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-10
Source-Free Cluster Adaptation for Privacy-Preserving Machinery Fault Diagnosis
Unsupervised domain adaptation (UDA) has been widely exploited for machinery fault diagnosis (MFD). However, existing UDA approaches always require direct access and sharing to the labeled source domain, raising privacy concerns. In this paper, a practical and challenging scenario, source-free UDA (SFUDA), is considered for privacy-preserving MFD. In SFUDA, only a pre-trained source model is provided for the unlabeled target domain, and the source data is inaccessible during adaptation. A novel SFUDA approach, namely source-free cluster adaptation (SF-CA), is proposed, which consists of source domain generalization (SDG) and target model adaptation (TMA). SDG aims to obtain a well-generalized source model for TMA. Specifically, the adaptive R-drop is proposed for SDG, where an entropy-aware weighted consistency training strategy is introduced to regularize dropout. TMA enforces cluster assumption for adaptation, where the structure exploration regularizations are proposed to learn structural information. Besides, selective self-training is introduced to alleviate model collapse during TMA. Finally, discriminative and tight-clustered target features can be obtained for SFUDA. Extensive experiments are conducted on three public datasets and one practical dataset. The experimental results show the effectiveness of SF-CA for privacy-preserving MFD and the feasibility of cluster assumption enforcement for SFUDA.
- 7 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-10
Source-Free Cluster Adaptation for Privacy-Preserving Machinery Fault Diagnosis
Unsupervised domain adaptation (UDA) has been widely exploited for machinery fault diagnosis (MFD). However, existing UDA approaches always require direct access and sharing to the labeled source domain, raising privacy concerns. In this paper, a practical and challenging scenario, source-free UDA (SFUDA), is considered for privacy-preserving MFD. In SFUDA, only a pre-trained source model is provided for the unlabeled target domain, and the source data is inaccessible during adaptation. A novel SFUDA approach, namely source-free cluster adaptation (SF-CA), is proposed, which consists of source domain generalization (SDG) and target model adaptation (TMA). SDG aims to obtain a well-generalized source model for TMA. Specifically, the adaptive R-drop is proposed for SDG, where an entropy-aware weighted consistency training strategy is introduced to regularize dropout. TMA enforces cluster assumption for adaptation, where the structure exploration regularizations are proposed to learn structural information. Besides, selective self-training is introduced to alleviate model collapse during TMA. Finally, discriminative and tight-clustered target features can be obtained for SFUDA. Extensive experiments are conducted on three public datasets and one practical dataset. The experimental results show the effectiveness of SF-CA for privacy-preserving MFD and the feasibility of cluster assumption enforcement for SFUDA.
- 19 May 2021
- Published by IOP Publishing in Nanotechnology
- Vol. 32 (32) , 325205
Controllable high-performance memristors based on 2D Fe 2 GeTe 3 oxide for biological synapse imitation
Memristors are an important component of the next-generation artificial neural network, high computing systems, etc. In the past, two-dimensional materials based memristors have achieved a high performance and low power consumption, though one at the cost of the other. Furthermore, their performance can not be modulated frequently once their structures are fixed, which remains the bottleneck in the development. Herein, a series of forming free memristors are fabricated with the same Cu/Fe 3 GeTe 2 oxide/Fe 3 GeTe 2 /Al structure, yet the On/Off ratio and set voltage is modulated continuously by varying the oxidation time during fabrication. With an optimal oxidation time, a large On/Off ratio (1.58 × 10 3 ) and low set voltage (0.74 V) is achieved in a single device. The formation and rapture of Al conductive filaments are found to be responsible for the memristors, and the filaments density and the cross-section area increase with the increase of current compliance, which achieves a higher On/Off ratio. The memristor can imitate basic biological synaptic functions using voltage pulses, demonstrating the potential for low-power consuming neuromorphic computing applications.
- 19 May 2021
- Published by IOP Publishing in Nanotechnology
- Vol. 32 (32) , 325205
Controllable high-performance memristors based on 2D Fe 2 GeTe 3 oxide for biological synapse imitation
Memristors are an important component of the next-generation artificial neural network, high computing systems, etc. In the past, two-dimensional materials based memristors have achieved a high performance and low power consumption, though one at the cost of the other. Furthermore, their performance can not be modulated frequently once their structures are fixed, which remains the bottleneck in the development. Herein, a series of forming free memristors are fabricated with the same Cu/Fe 3 GeTe 2 oxide/Fe 3 GeTe 2 /Al structure, yet the On/Off ratio and set voltage is modulated continuously by varying the oxidation time during fabrication. With an optimal oxidation time, a large On/Off ratio (1.58 × 10 3 ) and low set voltage (0.74 V) is achieved in a single device. The formation and rapture of Al conductive filaments are found to be responsible for the memristors, and the filaments density and the cross-section area increase with the increase of current compliance, which achieves a higher On/Off ratio. The memristor can imitate basic biological synaptic functions using voltage pulses, demonstrating the potential for low-power consuming neuromorphic computing applications.
- 27 June 2014
- Published by Institute of Advanced Engineering and Science in TELKOMNIKA Indonesian Journal of Electrical Engineering
- Vol. 12 (9) , 6725-6731
Intelligent Train Operation Models Based on Ensemble Regression Trees
Traditional control algorithms in Automatic Train Operation (ATO) system have some drawbacks, such as high energy consumption and low riding comfort. Combined with data mining methods and driving experience, two Intelligent Train Operation (ITO) models for the subway train control are proposed. Firstly the training data set was sorted out and sieved out from the real train operation data set by drivers in Beijing subway line Yizhuang to establish the standard database. By using Classification and Regression Trees (CART) algorithm and Bagging ensemble learning method which base on CART algorithm, two ITO models are dug out to represent the output of controller with limited speed, running time and gradient. In the train control simulation platform, ITO models were compared with the traditional PID (Proportional Integral Derivative) control algorithm of ATO systems. The simulation results indicate the proposed ITO models are better than PID control in energy consumption, riding comfort and switching times of controller’s output. Furthermore, the ITO model with bagging ensemble learning method is better especially in energy consumption and riding comfort.
- 27 June 2014
- Published by Institute of Advanced Engineering and Science in TELKOMNIKA Indonesian Journal of Electrical Engineering
- Vol. 12 (9) , 6725-6731
Intelligent Train Operation Models Based on Ensemble Regression Trees
Traditional control algorithms in Automatic Train Operation (ATO) system have some drawbacks, such as high energy consumption and low riding comfort. Combined with data mining methods and driving experience, two Intelligent Train Operation (ITO) models for the subway train control are proposed. Firstly the training data set was sorted out and sieved out from the real train operation data set by drivers in Beijing subway line Yizhuang to establish the standard database. By using Classification and Regression Trees (CART) algorithm and Bagging ensemble learning method which base on CART algorithm, two ITO models are dug out to represent the output of controller with limited speed, running time and gradient. In the train control simulation platform, ITO models were compared with the traditional PID (Proportional Integral Derivative) control algorithm of ATO systems. The simulation results indicate the proposed ITO models are better than PID control in energy consumption, riding comfort and switching times of controller’s output. Furthermore, the ITO model with bagging ensemble learning method is better especially in energy consumption and riding comfort.