Guangyu Li
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
Guangyu Li has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Fan Song
School of Optometry, Hong Kong Polytechnic University, Hong Kong, China (Hong Kong)
2 shared publications - Jian Zhang
State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, Guangdong, China
2 shared publications - Mengying Lai
School of Optometry, Hong Kong Polytechnic University, Hong Kong, China (Hong Kong)
2 shared publications - Mingge Li
Department of Ophthalmology, Shenzhen Nanshan People's Hospital, Shenzhen, Guangdong, China
2 shared publications - Mingguang He
School of Optometry, The Hong Kong Polytechnic University, Hong Kong SAR, China
2 shared publications - Yanxian Chen
School of Optometry, The Hong Kong Polytechnic University, Hong Kong, China
2 shared publications
Latest Publications
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-9
PIE-PS: Photometric Stereo from Physical Irradiance Event Streams
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-9
Open AccessPIE-PS: Photometric Stereo from Physical Irradiance Event Streams
- 27 July 2026
- Published by AIP Publishing in Applied Physics Letters
- Vol. 129 (4)
Volatile and nonvolatile resistive switching reconfiguration in CeO2-based memristor
The era of big data and the rapid development of artificial intelligence pose urgent demand on reconfigurability of memristive devices. Here, we demonstrate that a Pt/CeO2/V memristor can exhibit both nonvolatile and volatile characteristics through voltage polarity control and operation sequence. Nonvolatile switching is obtained by applying a positive sweep followed by a negative sweep, whereas volatile switching arises from consecutive positive sweeps. Notably, the device shows excellent performance in both operation modes. As a nonvolatile memory, it exhibits low coefficient of variation of set/reset voltage (17%), robust retention time (∼105 s), and stable on/off ratio (∼10). As a volatile memory, it displays low operation voltage (∼0.9 V), ultrahigh uniformity (∼1%), and high selectivity over 102. It is supposed that these behaviors can be attributed to polarity-dependent filament dynamics: negative bias induces filament rupture in the CeO2 layer, while positive bias governs filament formation and rupture in the VOx layer. This study is helpful for the development of CeO2-based memristive devices with configurable functionalities.
- 27 July 2026
- Published by AIP Publishing in Applied Physics Letters
- Vol. 129 (4)
Volatile and nonvolatile resistive switching reconfiguration in CeO2-based memristor
The era of big data and the rapid development of artificial intelligence pose urgent demand on reconfigurability of memristive devices. Here, we demonstrate that a Pt/CeO2/V memristor can exhibit both nonvolatile and volatile characteristics through voltage polarity control and operation sequence. Nonvolatile switching is obtained by applying a positive sweep followed by a negative sweep, whereas volatile switching arises from consecutive positive sweeps. Notably, the device shows excellent performance in both operation modes. As a nonvolatile memory, it exhibits low coefficient of variation of set/reset voltage (17%), robust retention time (∼105 s), and stable on/off ratio (∼10). As a volatile memory, it displays low operation voltage (∼0.9 V), ultrahigh uniformity (∼1%), and high selectivity over 102. It is supposed that these behaviors can be attributed to polarity-dependent filament dynamics: negative bias induces filament rupture in the CeO2 layer, while positive bias governs filament formation and rupture in the VOx layer. This study is helpful for the development of CeO2-based memristive devices with configurable functionalities.
- Open Access
- 28 April 2026
- Published by JMIR Publications Inc. in JMIR Human Factors
- Vol. 13 (1) , e84479
Identifying Biomarkers Using a Portable, Home-Based Eye-Tracking System to Predict Short-Term Visual Fatigue Deterioration: Prospective Observational Feasibility Study
Background: The escalating prevalence of screen-related eye fatigue has become a health burden in the digital era worldwide, yet routine monitoring relies largely on subjective reports. This underscores the urgent need for clinically applicable, objective diagnostic solutions. Ocular metrics provide an objective method to assess computer vision syndrome, or asthenopia. Objective: This study aimed to develop and evaluate an integrated at-home system for predicting short-term deteriorated asthenopia using objective ocular metrics. This system classifies the short-term risk level for practical monitoring and automatically generates a session report that summarizes metrics to complement symptom-based evaluation. Methods: We developed EyeFatigue Tracker, an integrated at-home system delivered via a desktop app, comprising a head-mounted device to record binocular infrared eye videos, a deep learning (DL) model to extract ocular metrics, and a machine learning (ML) classifier to estimate asthenopia risk. The DL model, trained on an in-house dataset, segments the palpebral fissure, pupil, and iris from recorded videos to derive ocular metrics. To build the prediction model, participants were recruited to complete a 1-hour computer gameplay session. Changes in the Computer Vision Syndrome Questionnaire (CVS-Q) scores served as the primary outcome measure to classify participants into deteriorated and nondeteriorated asthenopia groups. Metrics showing significant between-group differences were used as inputs for four ML models, including support vector machine (SVM), decision tree, extreme gradient boosting (XGBoost), and random forest, to identify deteriorated asthenopia. Model performance was evaluated with fivefold cross-validation. Results: This study enrolled 38 participants aged 19-31 (mean 24.8, SD 3.11) years. Following visual tasks, participants’ CVS-Q scores were higher compared to baseline values (mean 9.21, SD 4.57, vs mean 6.76, SD 3.76; P<.001). Alongside the critical flicker fusion frequency (CFF), nine key features were selected as predictive indicators, with the top five reflecting fissure length variability (variance, coefficient of variation, and SD), average blink duration, and pupil size variability (coefficient of variation). Most ML models exhibited high discriminative ability, with the random forest achieving the best overall performance (mean accuracy 0.720, SD 0.035; mean area under the receiver operating characteristic curve 0.850, 95% CI 0.830-0.860). Conclusions: The findings highlight the potential of objective indicators in identifying individuals at risk for asthenopia following computer gameplay. The ML models using ocular biomarkers identified in this study achieved plausible discriminative ability in detecting deteriorated asthenopia. EyeFatigue Tracker functions as an integrated, at-home system that produces a risk level prediction and a concise session report, supporting early detection and informing preventive care in real-world settings.
- 28 April 2026
- Published by JMIR Publications Inc. in JMIR Human Factors
- Vol. 13 (1) , e84479
Open AccessIdentifying Biomarkers Using a Portable, Home-Based Eye-Tracking System to Predict Short-Term Visual Fatigue Deterioration: Prospective Observational Feasibility Study
Background: The escalating prevalence of screen-related eye fatigue has become a health burden in the digital era worldwide, yet routine monitoring relies largely on subjective reports. This underscores the urgent need for clinically applicable, objective diagnostic solutions. Ocular metrics provide an objective method to assess computer vision syndrome, or asthenopia. Objective: This study aimed to develop and evaluate an integrated at-home system for predicting short-term deteriorated asthenopia using objective ocular metrics. This system classifies the short-term risk level for practical monitoring and automatically generates a session report that summarizes metrics to complement symptom-based evaluation. Methods: We developed EyeFatigue Tracker, an integrated at-home system delivered via a desktop app, comprising a head-mounted device to record binocular infrared eye videos, a deep learning (DL) model to extract ocular metrics, and a machine learning (ML) classifier to estimate asthenopia risk. The DL model, trained on an in-house dataset, segments the palpebral fissure, pupil, and iris from recorded videos to derive ocular metrics. To build the prediction model, participants were recruited to complete a 1-hour computer gameplay session. Changes in the Computer Vision Syndrome Questionnaire (CVS-Q) scores served as the primary outcome measure to classify participants into deteriorated and nondeteriorated asthenopia groups. Metrics showing significant between-group differences were used as inputs for four ML models, including support vector machine (SVM), decision tree, extreme gradient boosting (XGBoost), and random forest, to identify deteriorated asthenopia. Model performance was evaluated with fivefold cross-validation. Results: This study enrolled 38 participants aged 19-31 (mean 24.8, SD 3.11) years. Following visual tasks, participants’ CVS-Q scores were higher compared to baseline values (mean 9.21, SD 4.57, vs mean 6.76, SD 3.76; P<.001). Alongside the critical flicker fusion frequency (CFF), nine key features were selected as predictive indicators, with the top five reflecting fissure length variability (variance, coefficient of variation, and SD), average blink duration, and pupil size variability (coefficient of variation). Most ML models exhibited high discriminative ability, with the random forest achieving the best overall performance (mean accuracy 0.720, SD 0.035; mean area under the receiver operating characteristic curve 0.850, 95% CI 0.830-0.860). Conclusions: The findings highlight the potential of objective indicators in identifying individuals at risk for asthenopia following computer gameplay. The ML models using ocular biomarkers identified in this study achieved plausible discriminative ability in detecting deteriorated asthenopia. EyeFatigue Tracker functions as an integrated, at-home system that produces a risk level prediction and a concise session report, supporting early detection and informing preventive care in real-world settings.
- Open Access
- 20 September 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Identifying Biomarkers Using a Portable, Home-Based Eye-Tracking System to Predict Short-Term Visual Fatigue Deterioration: Prospective Observational Feasibility Study (Preprint)
BACKGROUND The escalating prevalence of screen-related eye fatigue has become a health burden in the digital era worldwide, yet routine monitoring relies largely on subjective reports. This underscores the urgent need for clinically applicable, objective diagnostic solutions. Ocular metrics provide an objective method to assess computer vision syndrome, or asthenopia. OBJECTIVE This study aimed to develop and evaluate an integrated at-home system for predicting short-term deteriorated asthenopia using objective ocular metrics. This system classifies the short-term risk level for practical monitoring and automatically generates a session report that summarizes metrics to complement symptom-based evaluation. METHODS We developed EyeFatigue Tracker, an integrated at-home system delivered via a desktop app, comprising a head-mounted device to record binocular infrared eye videos, a deep learning (DL) model to extract ocular metrics, and a machine learning (ML) classifier to estimate asthenopia risk. The DL model, trained on an in-house dataset, segments the palpebral fissure, pupil, and iris from recorded videos to derive ocular metrics. To build the prediction model, participants were recruited to complete a 1-hour computer gameplay session. Changes in the Computer Vision Syndrome Questionnaire (CVS-Q) scores served as the primary outcome measure to classify participants into deteriorated and nondeteriorated asthenopia groups. Metrics showing significant between-group differences were used as inputs for four ML models, including support vector machine (SVM), decision tree, extreme gradient boosting (XGBoost), and random forest, to identify deteriorated asthenopia. Model performance was evaluated with fivefold cross-validation. RESULTS This study enrolled 38 participants aged 19-31 (mean 24.8, SD 3.11) years. Following visual tasks, participants’ CVS-Q scores were higher compared to baseline values (mean 9.21, SD 4.57, vs mean 6.76, SD 3.76; P<.001). Alongside the critical flicker fusion frequency (CFF), nine key features were selected as predictive indicators, with the top five reflecting fissure length variability (variance, coefficient of variation, and SD), average blink duration, and pupil size variability (coefficient of variation). Most ML models exhibited high discriminative ability, with the random forest achieving the best overall performance (mean accuracy 0.720, SD 0.035; mean area under the receiver operating characteristic curve 0.850, 95% CI 0.830-0.860). CONCLUSIONS The findings highlight the potential of objective indicators in identifying individuals at risk for asthenopia following computer gameplay. The ML models using ocular biomarkers identified in this study achieved plausible discriminative ability in detecting deteriorated asthenopia. EyeFatigue Tracker functions as an integrated, at-home system that produces a risk level prediction and a concise session report, supporting early detection and informing preventive care in real-world settings.
- 20 September 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessIdentifying Biomarkers Using a Portable, Home-Based Eye-Tracking System to Predict Short-Term Visual Fatigue Deterioration: Prospective Observational Feasibility Study (Preprint)
BACKGROUND The escalating prevalence of screen-related eye fatigue has become a health burden in the digital era worldwide, yet routine monitoring relies largely on subjective reports. This underscores the urgent need for clinically applicable, objective diagnostic solutions. Ocular metrics provide an objective method to assess computer vision syndrome, or asthenopia. OBJECTIVE This study aimed to develop and evaluate an integrated at-home system for predicting short-term deteriorated asthenopia using objective ocular metrics. This system classifies the short-term risk level for practical monitoring and automatically generates a session report that summarizes metrics to complement symptom-based evaluation. METHODS We developed EyeFatigue Tracker, an integrated at-home system delivered via a desktop app, comprising a head-mounted device to record binocular infrared eye videos, a deep learning (DL) model to extract ocular metrics, and a machine learning (ML) classifier to estimate asthenopia risk. The DL model, trained on an in-house dataset, segments the palpebral fissure, pupil, and iris from recorded videos to derive ocular metrics. To build the prediction model, participants were recruited to complete a 1-hour computer gameplay session. Changes in the Computer Vision Syndrome Questionnaire (CVS-Q) scores served as the primary outcome measure to classify participants into deteriorated and nondeteriorated asthenopia groups. Metrics showing significant between-group differences were used as inputs for four ML models, including support vector machine (SVM), decision tree, extreme gradient boosting (XGBoost), and random forest, to identify deteriorated asthenopia. Model performance was evaluated with fivefold cross-validation. RESULTS This study enrolled 38 participants aged 19-31 (mean 24.8, SD 3.11) years. Following visual tasks, participants’ CVS-Q scores were higher compared to baseline values (mean 9.21, SD 4.57, vs mean 6.76, SD 3.76; P<.001). Alongside the critical flicker fusion frequency (CFF), nine key features were selected as predictive indicators, with the top five reflecting fissure length variability (variance, coefficient of variation, and SD), average blink duration, and pupil size variability (coefficient of variation). Most ML models exhibited high discriminative ability, with the random forest achieving the best overall performance (mean accuracy 0.720, SD 0.035; mean area under the receiver operating characteristic curve 0.850, 95% CI 0.830-0.860). CONCLUSIONS The findings highlight the potential of objective indicators in identifying individuals at risk for asthenopia following computer gameplay. The ML models using ocular biomarkers identified in this study achieved plausible discriminative ability in detecting deteriorated asthenopia. EyeFatigue Tracker functions as an integrated, at-home system that produces a risk level prediction and a concise session report, supporting early detection and informing preventive care in real-world settings.
- Top Cited PapersOpen Access
- 2 February 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-14
Hybrid neural networks for continual learning inspired by corticohippocampal circuits
Current artificial systems suffer from catastrophic forgetting during continual learning, a limitation absent in biological systems. Biological mechanisms leverage the dual representation of specific and generalized memories within corticohippocampal circuits to facilitate lifelong learning. Inspired by this, we develop a corticohippocampal circuits-based hybrid neural network (CH-HNN) that emulates these dual representations, significantly mitigating catastrophic forgetting in both task-incremental and class-incremental learning scenarios. Our CH-HNNs incorporate artificial neural networks and spiking neural networks, leveraging prior knowledge to facilitate new concept learning through episode inference, and offering insights into the neural functions of both feedforward and feedback loops within corticohippocampal circuits. Crucially, CH-HNN operates as a task-agnostic system without increasing memory demands, demonstrating adaptability and robustness in real-world applications. Coupled with the low power consumption inherent to SNNs, our model represents the potential for energy-efficient, continual learning in dynamic environments.
- 2 February 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-14
Top Cited PapersOpen AccessHybrid neural networks for continual learning inspired by corticohippocampal circuits
Current artificial systems suffer from catastrophic forgetting during continual learning, a limitation absent in biological systems. Biological mechanisms leverage the dual representation of specific and generalized memories within corticohippocampal circuits to facilitate lifelong learning. Inspired by this, we develop a corticohippocampal circuits-based hybrid neural network (CH-HNN) that emulates these dual representations, significantly mitigating catastrophic forgetting in both task-incremental and class-incremental learning scenarios. Our CH-HNNs incorporate artificial neural networks and spiking neural networks, leveraging prior knowledge to facilitate new concept learning through episode inference, and offering insights into the neural functions of both feedforward and feedback loops within corticohippocampal circuits. Crucially, CH-HNN operates as a task-agnostic system without increasing memory demands, demonstrating adaptability and robustness in real-world applications. Coupled with the low power consumption inherent to SNNs, our model represents the potential for energy-efficient, continual learning in dynamic environments.
- Open Access
- 21 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 32 (15344320) , 2078-2086
Deep Learning for Electromyographic Lower-Limb Motion Signal Classification Using Residual Learning
Electromyographic (EMG) signals have gained popularity for controlling prostheses and exoskeletons, particularly in the field of upper limbs for stroke patients. However, there is a lack of research in the lower limb area, and standardized open-source datasets of lower limb EMG signals, especially recording data of Asian race features, are scarce. Additionally, deep learning algorithms are rarely used for human motion intention recognition based on EMG, especially in the lower limb area. In response to these gaps, we present an open-source benchmark dataset of lower limb EMG with Asian race characteristics and large data volume, the JJ dataset, which includes approximately 13,350 clean EMG segments of 10 gait phases from 15 people. This is the first dataset of its kind to include the nine main muscles of human gait when walking. We used the processed time-domain signal as input and adjusted ResNet-18 as the classification tool. Our research explores and compares multiple key issues in this area, including the comparison of sliding time window method and other preprocessing methods, comparison of time-domain and frequency-domain signal processing effects, cross-subject motion recognition accuracy, and the possibility of using thigh and calf muscles in amputees. Our experiments demonstrate that the adjusted ResNet can achieve significant classification accuracy, with an average accuracy rate of 95.34% for human gait phases. Our research provides a valuable resource for future studies in this area and demonstrates the potential for ResNet as a robust and effective method for lower limb human motion intention pattern recognition.
- 21 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 32 (15344320) , 2078-2086
Open AccessDeep Learning for Electromyographic Lower-Limb Motion Signal Classification Using Residual Learning
Electromyographic (EMG) signals have gained popularity for controlling prostheses and exoskeletons, particularly in the field of upper limbs for stroke patients. However, there is a lack of research in the lower limb area, and standardized open-source datasets of lower limb EMG signals, especially recording data of Asian race features, are scarce. Additionally, deep learning algorithms are rarely used for human motion intention recognition based on EMG, especially in the lower limb area. In response to these gaps, we present an open-source benchmark dataset of lower limb EMG with Asian race characteristics and large data volume, the JJ dataset, which includes approximately 13,350 clean EMG segments of 10 gait phases from 15 people. This is the first dataset of its kind to include the nine main muscles of human gait when walking. We used the processed time-domain signal as input and adjusted ResNet-18 as the classification tool. Our research explores and compares multiple key issues in this area, including the comparison of sliding time window method and other preprocessing methods, comparison of time-domain and frequency-domain signal processing effects, cross-subject motion recognition accuracy, and the possibility of using thigh and calf muscles in amputees. Our experiments demonstrate that the adjusted ResNet can achieve significant classification accuracy, with an average accuracy rate of 95.34% for human gait phases. Our research provides a valuable resource for future studies in this area and demonstrates the potential for ResNet as a robust and effective method for lower limb human motion intention pattern recognition.