Yi Gu
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
Yi Gu has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
Alex Ce Zhang
Department of Electrical and Computer Engineering, University of California San Diego, La Jolla, California 92093, USA
4 shared publications- Lu Wang
School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, China
4 shared publications - Xiaoqing Gu
School of Computer Science and Artificial Intelligence, Changzhou University, Changzhou, China
4 shared publications Yuanyuan Han
Department of Electrical and Computer Engineering, University of California San Diego, La Jolla, California 92093, USA
4 shared publications- Yizhang Jiang
Jiangnan University
3 shared publications - Yu-Hwa Lo
University of California, San Diego
3 shared publications
Latest Publications
- 19 July 2026
- Published by Springer Nature
- p. 158-168
MCT-Net: Multi-task Clustering Transformer for Lightweight 3D Abdominal Multi-organ CT Segmentation
- 19 July 2026
- Published by Springer Nature
- p. 158-168
MCT-Net: Multi-task Clustering Transformer for Lightweight 3D Abdominal Multi-organ CT Segmentation
- 17 October 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 25  (10)
A CNN-TRANSFORMER-BASED BRAIN-COMPUTER INTERFACE MODEL FOR DIGITAL ART THERAPY IN DEPRESSION INTERVENTION AMONG COLLEGE STUDENTS
Targeting the depression-prone college student population, this study proposes an intelligent intervention system integrating brain–computer interface (BCI) technology with digital art therapy. Based on multi-channel electroencephalogram (EEG) signals, a hierarchical feature extraction and classification model is designed, consisting of a Time–Frequency Encoding module (TFE), a Multiscale Temporal Modeling module (MSTM), and an Emotion Discrimination module (EDM). The TFE module extracts local time–frequency features of EEG signals using multi-scale convolution and pooling structures. The MSTM module employs a dual-branch architecture combining coarse- and fine-grained convolutions with a Transformer mechanism to capture both local details and global temporal dependencies, achieving multiscale temporal feature modeling. The EDM module performs multi-class emotion classification using a fully connected network. Based on the recognized emotional states, an interactive digital art feedback module is developed, mapping real-time emotion states to artistic visual images to promote emotional regulation and self-expression. Experimental validation across multiple participants demonstrates that the proposed model achieves high accuracy and good cross-subject generalization in multi-class emotion recognition tasks. Moreover, the digital art feedback module shows a positive effect in improving participants’ psychological states. The proposed intervention framework combines real-time processing, hierarchical feature modeling, and cross-subject adaptability, offering a novel technological solution and application reference for digital mental health interventions in university settings.
- 17 October 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 25  (10)
A CNN-TRANSFORMER-BASED BRAIN-COMPUTER INTERFACE MODEL FOR DIGITAL ART THERAPY IN DEPRESSION INTERVENTION AMONG COLLEGE STUDENTS
Targeting the depression-prone college student population, this study proposes an intelligent intervention system integrating brain–computer interface (BCI) technology with digital art therapy. Based on multi-channel electroencephalogram (EEG) signals, a hierarchical feature extraction and classification model is designed, consisting of a Time–Frequency Encoding module (TFE), a Multiscale Temporal Modeling module (MSTM), and an Emotion Discrimination module (EDM). The TFE module extracts local time–frequency features of EEG signals using multi-scale convolution and pooling structures. The MSTM module employs a dual-branch architecture combining coarse- and fine-grained convolutions with a Transformer mechanism to capture both local details and global temporal dependencies, achieving multiscale temporal feature modeling. The EDM module performs multi-class emotion classification using a fully connected network. Based on the recognized emotional states, an interactive digital art feedback module is developed, mapping real-time emotion states to artistic visual images to promote emotional regulation and self-expression. Experimental validation across multiple participants demonstrates that the proposed model achieves high accuracy and good cross-subject generalization in multi-class emotion recognition tasks. Moreover, the digital art feedback module shows a positive effect in improving participants’ psychological states. The proposed intervention framework combines real-time processing, hierarchical feature modeling, and cross-subject adaptability, offering a novel technological solution and application reference for digital mental health interventions in university settings.
- 20 July 2025
- Published by Springer Nature
- p. 77-89
An Untrained Single Shot Image Reconstruction Deep Neural Network for Compressive Sensing
- 20 July 2025
- Published by Springer Nature
- p. 77-89
An Untrained Single Shot Image Reconstruction Deep Neural Network for Compressive Sensing
- 18 March 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 25  (02)
A COMPRESSED SENSING RECONSTRUCTION METHOD BASED ON UNTRAINED CONVOLUTIONAL NEURAL NETWORK FOR MEDICAL IMAGING
Medical imaging holds significant importance in disease diagnosis and surgical preparation, necessitating higher standards for the storage and transmission of medical images. We propose a method for compressed sensing reconstruction of medical imaging based on untrained convolutional neural networks. This method does not require any training data, offering vast potential in the field of medical imaging. We employ the generator model of Deep Convolutional Generative Adversarial Networks (DCGAN) as the structure for the untrained convolutional neural network. We utilized the Musculoskeletal Radiographs (MURA) and Open Access Series of Imaging Studies (OASIS) datasets to demonstrate that our method significantly improved the reconstruction effect in medical imaging, regardless of whether the sampling rate was high or low.
- 18 March 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 25  (02)
A COMPRESSED SENSING RECONSTRUCTION METHOD BASED ON UNTRAINED CONVOLUTIONAL NEURAL NETWORK FOR MEDICAL IMAGING
Medical imaging holds significant importance in disease diagnosis and surgical preparation, necessitating higher standards for the storage and transmission of medical images. We propose a method for compressed sensing reconstruction of medical imaging based on untrained convolutional neural networks. This method does not require any training data, offering vast potential in the field of medical imaging. We employ the generator model of Deep Convolutional Generative Adversarial Networks (DCGAN) as the structure for the untrained convolutional neural network. We utilized the Musculoskeletal Radiographs (MURA) and Open Access Series of Imaging Studies (OASIS) datasets to demonstrate that our method significantly improved the reconstruction effect in medical imaging, regardless of whether the sampling rate was high or low.
- 16 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71  (2) , 7170-7179
Driving Behavior Safety Assessment in Edge Computing Using Multitask Discriminative TS Fuzzy Model
Driving behavior safety analysis in edge computing involves the real-time collection and processing of vehicle driving data. Fuzzy models have advantages in analyzing driving behavior due to their excellent ability to handle uncertain information. This study develops a multitask discriminative Takagi-Sugeno fuzzy model (MD-TS-FM) for driving behavior safety assessment. To comprehensively analyze the correlation and differences in multitask driving behaviors, the designed consequent part consists of two parts: task-shared part reflects the shared intrinsic structure information that are consistent across different driving tasks, and task-specific part reflects distinct characteristics and variations specific to each task, allowing the model to address the unique aspects of different driving behaviors. Accordingly, the task-shared consequent part is characterized by low-rank property, which mines global structural information within the fuzzy space; whereas the task-specific consequent part exhibits sparsity, which removes non-discriminative and irrelevant information. This sparsity ensures that the model focuses on the most critical features for each specific task. Furthermore, a discriminative diversity term is introduced to enhance the diversity between tasks, which explores the consistent information of task-shared consequents while reducing the overlap of task-specific consequents. Experimental results indicate that the MD-TS-FM model can be effectively applied to driving behavior safety assessment.
- 16 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71  (2) , 7170-7179
Driving Behavior Safety Assessment in Edge Computing Using Multitask Discriminative TS Fuzzy Model
Driving behavior safety analysis in edge computing involves the real-time collection and processing of vehicle driving data. Fuzzy models have advantages in analyzing driving behavior due to their excellent ability to handle uncertain information. This study develops a multitask discriminative Takagi-Sugeno fuzzy model (MD-TS-FM) for driving behavior safety assessment. To comprehensively analyze the correlation and differences in multitask driving behaviors, the designed consequent part consists of two parts: task-shared part reflects the shared intrinsic structure information that are consistent across different driving tasks, and task-specific part reflects distinct characteristics and variations specific to each task, allowing the model to address the unique aspects of different driving behaviors. Accordingly, the task-shared consequent part is characterized by low-rank property, which mines global structural information within the fuzzy space; whereas the task-specific consequent part exhibits sparsity, which removes non-discriminative and irrelevant information. This sparsity ensures that the model focuses on the most critical features for each specific task. Furthermore, a discriminative diversity term is introduced to enhance the diversity between tasks, which explores the consistent information of task-shared consequents while reducing the overlap of task-specific consequents. Experimental results indicate that the MD-TS-FM model can be effectively applied to driving behavior safety assessment.
- 19 November 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 24  (09)
A STUDENT SENTIMENT ANALYSIS METHOD BASED ON MULTIMODAL DEEP LEARNING
Based on electroencephalography (EEG) and video data, we propose a multimodal affective analysis approach in this study to examine the affective states of university students. This method is based on the findings of this investigation. The EEG signals and video data were obtained from 50 college students experiencing various emotional states, and then they were processed in great detail. The EEG signals are pre-processed to extract their multi-view characteristics. Additionally, the video data were processed by frame extraction, face detection, and convolutional neural network (CNN) operations to extract features. We take a feature splicing strategy to merge EEG and video data to produce a time series input to realize the fusion of multimodal features. This allows us to realize the fusion of multimodal features. In addition, we developed and trained a model for the classification of emotional states based on a long short-term memory network (LSTM). With the help of cross-validation, the experiments were carried out by dividing the dataset into a training set and a test set. The model’s performance was evaluated with the help of four metrics: accuracy, precision, recall, and F1-score. When compared to the single-modal method of sentiment analysis, the results demonstrate that the multimodal approach, which combines EEG and video, demonstrates considerable advantages in terms of sentiment detection. Specifically, the accuracy obtained from the multimodal approach is significantly higher. As part of its investigation, the study also investigates the respective contributions of EEG and video aspects to emotion detection. It discovers that these features complement each other in a variety of emotional states and have the potential to improve the overall recognition results. The multimodal sentiment analysis method that is based on LSTM offers a high level of accuracy and robustness when it comes to recognizing the affective states of college students. This is especially essential for enhancing the quality of education and providing support for mental health.
- 19 November 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 24  (09)
A STUDENT SENTIMENT ANALYSIS METHOD BASED ON MULTIMODAL DEEP LEARNING
Based on electroencephalography (EEG) and video data, we propose a multimodal affective analysis approach in this study to examine the affective states of university students. This method is based on the findings of this investigation. The EEG signals and video data were obtained from 50 college students experiencing various emotional states, and then they were processed in great detail. The EEG signals are pre-processed to extract their multi-view characteristics. Additionally, the video data were processed by frame extraction, face detection, and convolutional neural network (CNN) operations to extract features. We take a feature splicing strategy to merge EEG and video data to produce a time series input to realize the fusion of multimodal features. This allows us to realize the fusion of multimodal features. In addition, we developed and trained a model for the classification of emotional states based on a long short-term memory network (LSTM). With the help of cross-validation, the experiments were carried out by dividing the dataset into a training set and a test set. The model’s performance was evaluated with the help of four metrics: accuracy, precision, recall, and F1-score. When compared to the single-modal method of sentiment analysis, the results demonstrate that the multimodal approach, which combines EEG and video, demonstrates considerable advantages in terms of sentiment detection. Specifically, the accuracy obtained from the multimodal approach is significantly higher. As part of its investigation, the study also investigates the respective contributions of EEG and video aspects to emotion detection. It discovers that these features complement each other in a variety of emotional states and have the potential to improve the overall recognition results. The multimodal sentiment analysis method that is based on LSTM offers a high level of accuracy and robustness when it comes to recognizing the affective states of college students. This is especially essential for enhancing the quality of education and providing support for mental health.
- 20 August 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 24  (08)
TAKAGI–SUGENO–KANG FUZZY SYSTEM MODELING BASED ON LOW-RANK SPARSE SUBSPACE LEARNING FOR MOTOR IMAGERY ELECTROENCEPHALOGRAM SIGNAL CLASSIFICATION
The classification of electroencephalogram (EEG) signals derived from motor imagery (MI) has always been a hot topic in the field of brain–computer interfaces. Due to its ability to handle the nonstationary and uncertain information contained in EEG signals, the Takagi–Sugeno–Kang fuzzy system (TSK-FS) has become an advantageous classification algorithm. To train a fuzzy system with strong discrimination capabilities from EEG data interspersed with redundant information, this paper proposes a TSK-FS modeling method based on low-rank sparse subspace learning (TSK-LSSL). This method focuses on consequent parameter learning, which transforms the traditional consequent parameter learning strategy into low-rank subspace and sparse subspace learning processes. Low-rank subspace learning is used to mine the global structural information of data and effectively reduce the number of fuzzy rules. During sparse subspace learning, [ ]-norm regularization is used to constrain the consequent parameters and causes the number of redundant consequent parameters to be zero, thereby simplifying the fuzzy rules. In addition, a local boundary term based on graph matrices is embedded into the objective function to mine the local structural information of the given data. TSK-LSSL simplifies the number of rules and the consequent part of the fuzzy rules. It exhibits good classification performance on two BCI Competition databases.
- 20 August 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Mechanics in Medicine and Biology
- Vol. 24  (08)
TAKAGI–SUGENO–KANG FUZZY SYSTEM MODELING BASED ON LOW-RANK SPARSE SUBSPACE LEARNING FOR MOTOR IMAGERY ELECTROENCEPHALOGRAM SIGNAL CLASSIFICATION
The classification of electroencephalogram (EEG) signals derived from motor imagery (MI) has always been a hot topic in the field of brain–computer interfaces. Due to its ability to handle the nonstationary and uncertain information contained in EEG signals, the Takagi–Sugeno–Kang fuzzy system (TSK-FS) has become an advantageous classification algorithm. To train a fuzzy system with strong discrimination capabilities from EEG data interspersed with redundant information, this paper proposes a TSK-FS modeling method based on low-rank sparse subspace learning (TSK-LSSL). This method focuses on consequent parameter learning, which transforms the traditional consequent parameter learning strategy into low-rank subspace and sparse subspace learning processes. Low-rank subspace learning is used to mine the global structural information of data and effectively reduce the number of fuzzy rules. During sparse subspace learning, [ ]-norm regularization is used to constrain the consequent parameters and causes the number of redundant consequent parameters to be zero, thereby simplifying the fuzzy rules. In addition, a local boundary term based on graph matrices is embedded into the objective function to mine the local structural information of the given data. TSK-LSSL simplifies the number of rules and the consequent part of the fuzzy rules. It exhibits good classification performance on two BCI Competition databases.
- 5 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11  (22) , 35752-35761
An IoT-Based Framework for Motion Planning of Connected Automated Vehicles at Signal-Free Traffic Intersections
Intersection management is an important problem to improve the traffic efficiency. A novel IoT framework (IoTF) is proposed to control vehicles at a signal-free intersection. An intersection includes the potential path conflict locations, which are represented by a timed Petri net in the IoTF. To ensure safety and efficiency, vehicles driving in the intersection should not occupy the same conflict location at the same time and should not block each other. In the cyberspace, the passing orders of the vehicles are calculated to control vehicles to pass through a signal-free intersection without any collisions and any deadlocks. The cost function aims to minimize the vehicles’ waiting time by optimizing their passing orders. Compared with existing works, 1) the IoTF makes full use of the finite spatial spacing of the intersection; 2) the IoTF are more efficient since they do not involve a complex nonlinear programming problem; 3) and the calculated control strategy aims to optimize the passing order of vehicles arriving at the intersection at each instant, i.e., it is a global optimum solution. The effectiveness of the developed algorithm is demonstrated by the simulation results.
- 5 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11  (22) , 35752-35761
An IoT-Based Framework for Motion Planning of Connected Automated Vehicles at Signal-Free Traffic Intersections
Intersection management is an important problem to improve the traffic efficiency. A novel IoT framework (IoTF) is proposed to control vehicles at a signal-free intersection. An intersection includes the potential path conflict locations, which are represented by a timed Petri net in the IoTF. To ensure safety and efficiency, vehicles driving in the intersection should not occupy the same conflict location at the same time and should not block each other. In the cyberspace, the passing orders of the vehicles are calculated to control vehicles to pass through a signal-free intersection without any collisions and any deadlocks. The cost function aims to minimize the vehicles’ waiting time by optimizing their passing orders. Compared with existing works, 1) the IoTF makes full use of the finite spatial spacing of the intersection; 2) the IoTF are more efficient since they do not involve a complex nonlinear programming problem; 3) and the calculated control strategy aims to optimize the passing order of vehicles arriving at the intersection at each instant, i.e., it is a global optimum solution. The effectiveness of the developed algorithm is demonstrated by the simulation results.
- Open Access
- 6 March 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
Transferred Low-rank Discriminative Sub-dictionary Learning Method for Information Identification and Understanding in IoT-based Natural Language Processing
As Internet of Things (IoT) technology progresses rapidly, there is an increasing demand for automatic identification and understanding of natural language data. However, data labeling requires large amounts of effort and cost. Most intelligent algorithms rely on the assumption of uniform distribution of data, which brings great challenge to IoT-based natural language processing. To solve this problem, this study develops a transferred low-rank discriminative sub-dictionary learning (TLDSL) method. The TLDSL method learns a shared subspace through the maximum mean discrepancy (MMD) strategy that minimizes the distribution difference of sparse coefficients between the source and target domains. By learning the common sub-dictionary of the two domains, TLDSL reveals the intrinsic connection and establishes a bridge between the two domains, thus completing the knowledge transfer. By introducing the sub-dictionary incoherence, TLDSL can avoid the atomic correlation between different sub-dictionaries. In addition, the sparse coefficients are constrained in low rank representation, which can improve the model discrimination ability while preserving the global data structure. Experiments show that the TLDSL method can be effectively performed on cross-domain text classification and handwritten digit recognition.
- 6 March 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
Open AccessTransferred Low-rank Discriminative Sub-dictionary Learning Method for Information Identification and Understanding in IoT-based Natural Language Processing
As Internet of Things (IoT) technology progresses rapidly, there is an increasing demand for automatic identification and understanding of natural language data. However, data labeling requires large amounts of effort and cost. Most intelligent algorithms rely on the assumption of uniform distribution of data, which brings great challenge to IoT-based natural language processing. To solve this problem, this study develops a transferred low-rank discriminative sub-dictionary learning (TLDSL) method. The TLDSL method learns a shared subspace through the maximum mean discrepancy (MMD) strategy that minimizes the distribution difference of sparse coefficients between the source and target domains. By learning the common sub-dictionary of the two domains, TLDSL reveals the intrinsic connection and establishes a bridge between the two domains, thus completing the knowledge transfer. By introducing the sub-dictionary incoherence, TLDSL can avoid the atomic correlation between different sub-dictionaries. In addition, the sparse coefficients are constrained in low rank representation, which can improve the model discrimination ability while preserving the global data structure. Experiments show that the TLDSL method can be effectively performed on cross-domain text classification and handwritten digit recognition.
- 11 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 24  (12) , 15807-15817
Transferable Takagi-Sugeno-Kang Fuzzy Classifier With Multi-Views for EEG-Based Driving Fatigue Recognition in Intelligent Transportation
The safety monitoring system of intelligent transportation provides driving fatigue warning and risk control. Electroencephalogram (EEG) signals can directly reflect the neuronal activity of the brain. The detection and early warning of driving fatigue using EEG signals has important practical significance. However, because of the non-stationarity and timeliness of EEG signals, the single feature detection method is significantly impacted by data distribution differences. In this paper, in the framework of multi-input multi-output (MIMO) Takagi-Sugeno-Kang (TSK) fuzzy system, transferable TSK fuzzy classifier with multi-views (T-TSK-MV) is developed for EEG-based driving fatigue recognition in intelligent transportation. First, in view-specific consequent parameter learning, the view-specific consequent regularizer is designed based on technologies of ridge regression, maximum mean discrepancy (MMD), and manifold regularization, which becomes the bridge to transfer the discriminative information from the related domain to the target domain. In addition, the $\ell_{2,1} $ -norm sparse constraint on consequent parameters is used to simplify fuzzy rules. Then multi-view learning is integrated into the consequent parameter learning, in which T-TSK-MV explores the view-shared consequent regularizer and adaptively assigns weights to each view. The $\ell_{2,1} $ -norm sparse constraint on view-shared consequent regularizer can effectively exploit the local structure of multi-view data. Finally, the fuzzy classifier is constructed on view-specific regularizers and view weights. The experiment on real-word datasets shows that the proposed fuzzy classifier can significantly improve the driving fatigue recognition performance.
- 11 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 24  (12) , 15807-15817
Transferable Takagi-Sugeno-Kang Fuzzy Classifier With Multi-Views for EEG-Based Driving Fatigue Recognition in Intelligent Transportation
The safety monitoring system of intelligent transportation provides driving fatigue warning and risk control. Electroencephalogram (EEG) signals can directly reflect the neuronal activity of the brain. The detection and early warning of driving fatigue using EEG signals has important practical significance. However, because of the non-stationarity and timeliness of EEG signals, the single feature detection method is significantly impacted by data distribution differences. In this paper, in the framework of multi-input multi-output (MIMO) Takagi-Sugeno-Kang (TSK) fuzzy system, transferable TSK fuzzy classifier with multi-views (T-TSK-MV) is developed for EEG-based driving fatigue recognition in intelligent transportation. First, in view-specific consequent parameter learning, the view-specific consequent regularizer is designed based on technologies of ridge regression, maximum mean discrepancy (MMD), and manifold regularization, which becomes the bridge to transfer the discriminative information from the related domain to the target domain. In addition, the $\ell_{2,1} $ -norm sparse constraint on consequent parameters is used to simplify fuzzy rules. Then multi-view learning is integrated into the consequent parameter learning, in which T-TSK-MV explores the view-shared consequent regularizer and adaptively assigns weights to each view. The $\ell_{2,1} $ -norm sparse constraint on view-shared consequent regularizer can effectively exploit the local structure of multi-view data. Finally, the fuzzy classifier is constructed on view-specific regularizers and view weights. The experiment on real-word datasets shows that the proposed fuzzy classifier can significantly improve the driving fatigue recognition performance.