Shih-Yeh Chen
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
Shih-Yeh Chen has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Chin-Feng Lai
Department of Engineering Science, National Cheng Kung University, Tainan, Taiwan
4 shared publications - Han-Chieh Chao
Department of Applied Informatics, Fo Guang University, Yilan 262307, Taiwan
2 shared publications - Hua-Yang Hsu
School of Electronic and Computer Engineering (SECE), Peking University Shenzhen Graduate School, Shenzhen 518055, China
2 shared publications Pei-Hsuan Lin
National Taichung University of Science and Technology, Taiwan
2 shared publications- Qi-Fong He
Taiwan Semiconductor Manufacturing Company Ltd., Hsinchu, Taiwan
2 shared publications - Wen Chi Chou
Taiwan Semiconductor Manufacturing Company, Hsinchu 300096, Taiwan
2 shared publications
Latest Publications
- 18 November 2026
- Published by Wiley in Transactions in GIS
- Vol. 30 (7)
Evaluating DEM‐Derived Hydrological Priors for U‐Net‐Based Sentinel‐1 SAR Flood Segmentation
Floods are among the most destructive natural hazards. Rapid and accurate flood extent mapping is essential for disaster response, risk assessment, and post‐disaster management. Synthetic Aperture Radar (SAR) imagery provides all‐weather and day‐and‐night observation capabilities and has become an important data source for flood monitoring. However, SAR backscatter alone does not explicitly represent terrain and drainage conditions. To examine whether complementary terrain information benefits flood mapping, this study systematically evaluates the effects of Digital Elevation Model (DEM)‐derived terrain and hydrological priors on SAR flood segmentation. Using the Sen1Floods11 dataset, this study constructs a three‐channel SAR baseline comprising normalized VV and VH polarizations and their pixel‐wise mean. Elevation and hydrological features derived from the GLO‐30 DEM, including flow accumulation (Flow), Topographic Wetness Index (TWI), and Height Above Nearest Drainage (HAND), are integrated as additional input channels. To ensure a fair comparison of different input features, the same U‐Net architecture, data split, and training settings are adopted to compare nine input configurations. Cross‐region generalization is further examined using a leave‐one‐region‐out (LORO) evaluation, and an additional DeepLabV3+ experiment is conducted to assess whether the contribution of Flow extends beyond U‐Net. Among the evaluated configurations, SAR+Flow achieved the highest mean IoU, increasing from 0.6800 for the SAR baseline to 0.6879. F1‐score increased from 0.8095 to 0.8151, and recall increased from 0.7958 to 0.8112. Under the LORO setting, SAR+Flow increased the region‐balanced mean IoU from 0.5680 to 0.5797, although improvements occurred in only 4 of the 10 held‐out regions. The DeepLabV3+ comparison also showed a small numerical improvement in mean IoU, increasing from 0.6555 to 0.6595 with the incorporation of Flow. Overall, the results suggest that Flow can provide useful complementary spatial information for SAR flood segmentation, but its contribution varies across geographic regions and model architectures.
- 18 November 2026
- Published by Wiley in Transactions in GIS
- Vol. 30 (7)
Evaluating DEM‐Derived Hydrological Priors for U‐Net‐Based Sentinel‐1 SAR Flood Segmentation
Floods are among the most destructive natural hazards. Rapid and accurate flood extent mapping is essential for disaster response, risk assessment, and post‐disaster management. Synthetic Aperture Radar (SAR) imagery provides all‐weather and day‐and‐night observation capabilities and has become an important data source for flood monitoring. However, SAR backscatter alone does not explicitly represent terrain and drainage conditions. To examine whether complementary terrain information benefits flood mapping, this study systematically evaluates the effects of Digital Elevation Model (DEM)‐derived terrain and hydrological priors on SAR flood segmentation. Using the Sen1Floods11 dataset, this study constructs a three‐channel SAR baseline comprising normalized VV and VH polarizations and their pixel‐wise mean. Elevation and hydrological features derived from the GLO‐30 DEM, including flow accumulation (Flow), Topographic Wetness Index (TWI), and Height Above Nearest Drainage (HAND), are integrated as additional input channels. To ensure a fair comparison of different input features, the same U‐Net architecture, data split, and training settings are adopted to compare nine input configurations. Cross‐region generalization is further examined using a leave‐one‐region‐out (LORO) evaluation, and an additional DeepLabV3+ experiment is conducted to assess whether the contribution of Flow extends beyond U‐Net. Among the evaluated configurations, SAR+Flow achieved the highest mean IoU, increasing from 0.6800 for the SAR baseline to 0.6879. F1‐score increased from 0.8095 to 0.8151, and recall increased from 0.7958 to 0.8112. Under the LORO setting, SAR+Flow increased the region‐balanced mean IoU from 0.5680 to 0.5797, although improvements occurred in only 4 of the 10 held‐out regions. The DeepLabV3+ comparison also showed a small numerical improvement in mean IoU, increasing from 0.6555 to 0.6595 with the incorporation of Flow. Overall, the results suggest that Flow can provide useful complementary spatial information for SAR flood segmentation, but its contribution varies across geographic regions and model architectures.
- Open Access
- 9 October 2026
- Published by Springer Nature in BMC Nursing
Preparing future nurses for person-centred mental health care through AI-assisted gamified learning: a mixed-methods study
- 9 October 2026
- Published by Springer Nature in BMC Nursing
Open AccessPreparing future nurses for person-centred mental health care through AI-assisted gamified learning: a mixed-methods study
- Open Access
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 15723-15735
TIPS: A Prompt Engineering Framework for Code Classification and Generation on Resource-Constrained Systems
Recent advances in AI-assisted programming have demonstrated the potential of Large Language Models (LLMs) in complex code classification and generation. However, their high computational demands and reliance on large-scale datasets limit deployment in resource-constrained environments such as educational tools, edge devices, and small-scale development settings. To address these challenges, this paper introduces the Template-Integrated Prompting System (TIPS), a modular prompt engineering framework designed for lightweight and efficient code analysis. TIPS integrates small-model training, semantically guided demonstration selection, and structured prompt templates to construct an interpretable reasoning pipeline that combines few-shot learning with Chain-of-Thought (CoT) reasoning. Experiments on a benchmark dataset of multithreaded Pthread programs show that TIPS achieves higher precision, recall, and F1-scores than traditional classifiers and prompt-based baselines, even under class imbalance. These results demonstrate that model effectiveness depends not only on scale but also on semantic guidance and structured reasoning. TIPS provides a scalable and interpretable solution, with potential applications in education, software development, and intelligent assistance systems, particularly in resource-constrained environments.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 15723-15735
Open AccessTIPS: A Prompt Engineering Framework for Code Classification and Generation on Resource-Constrained Systems
Recent advances in AI-assisted programming have demonstrated the potential of Large Language Models (LLMs) in complex code classification and generation. However, their high computational demands and reliance on large-scale datasets limit deployment in resource-constrained environments such as educational tools, edge devices, and small-scale development settings. To address these challenges, this paper introduces the Template-Integrated Prompting System (TIPS), a modular prompt engineering framework designed for lightweight and efficient code analysis. TIPS integrates small-model training, semantically guided demonstration selection, and structured prompt templates to construct an interpretable reasoning pipeline that combines few-shot learning with Chain-of-Thought (CoT) reasoning. Experiments on a benchmark dataset of multithreaded Pthread programs show that TIPS achieves higher precision, recall, and F1-scores than traditional classifiers and prompt-based baselines, even under class imbalance. These results demonstrate that model effectiveness depends not only on scale but also on semantic guidance and structured reasoning. TIPS provides a scalable and interpretable solution, with potential applications in education, software development, and intelligent assistance systems, particularly in resource-constrained environments.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
Cost-Aware Thresholding for Credit Card Fraud Detection: From Classification Metrics to Expected Error-Cost Minimization
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
Open AccessCost-Aware Thresholding for Credit Card Fraud Detection: From Classification Metrics to Expected Error-Cost Minimization
- Open Access
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 86 (2) , 1-19
A Cloud-Based Distributed System for Story Visualization Using Stable Diffusion
With the rapid development of generative artificial intelligence (GenAI), the task of story visualization, which transforms natural language narratives into coherent and consistent image sequences, has attracted growing research attention. However, existing methods still face limitations in balancing multi-frame character consistency and generation efficiency, which restricts their feasibility for large-scale practical applications. To address this issue, this study proposes a modular cloud-based distributed system built on Stable Diffusion. By separating the character generation and story generation processes, and integrating multi-feature control techniques, a caching mechanism, and an asynchronous task queue architecture, the system enhances generation efficiency and scalability. The experimental design includes both automated and human evaluations of character consistency, performance testing, and multi-node simulation. The results show that the proposed system outperforms the baseline model StoryGen in both CLIP-I and human evaluation metrics. In terms of performance, under the experimental environment of this study, dual-node deployment reduces average waiting time by approximately 19%, while the four-node simulation further reduces it by up to 65%. Overall, this study demonstrates the advantages of cloud-distributed GenAI in maintaining character consistency and reducing generation latency, highlighting its potential value in multi-user collaborative story visualization applications.
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 86 (2) , 1-19
Open AccessA Cloud-Based Distributed System for Story Visualization Using Stable Diffusion
With the rapid development of generative artificial intelligence (GenAI), the task of story visualization, which transforms natural language narratives into coherent and consistent image sequences, has attracted growing research attention. However, existing methods still face limitations in balancing multi-frame character consistency and generation efficiency, which restricts their feasibility for large-scale practical applications. To address this issue, this study proposes a modular cloud-based distributed system built on Stable Diffusion. By separating the character generation and story generation processes, and integrating multi-feature control techniques, a caching mechanism, and an asynchronous task queue architecture, the system enhances generation efficiency and scalability. The experimental design includes both automated and human evaluations of character consistency, performance testing, and multi-node simulation. The results show that the proposed system outperforms the baseline model StoryGen in both CLIP-I and human evaluation metrics. In terms of performance, under the experimental environment of this study, dual-node deployment reduces average waiting time by approximately 19%, while the four-node simulation further reduces it by up to 65%. Overall, this study demonstrates the advantages of cloud-distributed GenAI in maintaining character consistency and reducing generation latency, highlighting its potential value in multi-user collaborative story visualization applications.
- 10 December 2025
- Published by Wiley in Expert Systems
- Vol. 43 (1)
AI ‐Driven Intelligent Feedback System for Enhancing Self‐Assessment Accuracy in Higher Education Writing
With the rapid advancement of generative artificial intelligence, large language models (LLMs) have become increasingly integrated into education, particularly for automated formative feedback and writing assessment. This study introduces and evaluates an AI‐driven intelligent feedback system aimed at promoting sustainable and inclusive practices in higher education. It does so by delivering cost‐effective feedback in under‐resourced contexts and adaptive guidance for diverse learners. The system is built on transformer‐based models (BERT and RoBERTa) to support personalised writing evaluation and feedback generation. The system aims to improve students’ self‐assessment accuracy (SAA), a critical factor for self‐regulated learning, while addressing the challenge of delivering high‐quality feedback efficiently in under‐resourced contexts. A quasi‐experimental design was employed to examine the effects of LLM‐generated feedback (LLMF) on students’ SAA and to investigate how these effects vary by initial ability. Results indicated no significant group‐level difference in posttest SAA between the experimental and control groups. More importantly, interaction analysis revealed a significant moderating effect of Initial Self‐Assessment Accuracy (ISAA). Students with lower baseline accuracy benefited substantially from LLMF, while those with higher baseline SAA showed limited change. This compensatory effect highlights the potential of LLMF to reduce inequities in self‐regulated learning. These findings demonstrate the potential of AI‐driven feedback systems to cost‐effectively reduce calibration gaps and foster metacognitive development. By embedding adaptive and personalised mechanisms, such systems advance educational equity and promote scalable personalised learning. They also contribute to the broader agenda of intelligent and sustainable education.
- 10 December 2025
- Published by Wiley in Expert Systems
- Vol. 43 (1)
AI ‐Driven Intelligent Feedback System for Enhancing Self‐Assessment Accuracy in Higher Education Writing
With the rapid advancement of generative artificial intelligence, large language models (LLMs) have become increasingly integrated into education, particularly for automated formative feedback and writing assessment. This study introduces and evaluates an AI‐driven intelligent feedback system aimed at promoting sustainable and inclusive practices in higher education. It does so by delivering cost‐effective feedback in under‐resourced contexts and adaptive guidance for diverse learners. The system is built on transformer‐based models (BERT and RoBERTa) to support personalised writing evaluation and feedback generation. The system aims to improve students’ self‐assessment accuracy (SAA), a critical factor for self‐regulated learning, while addressing the challenge of delivering high‐quality feedback efficiently in under‐resourced contexts. A quasi‐experimental design was employed to examine the effects of LLM‐generated feedback (LLMF) on students’ SAA and to investigate how these effects vary by initial ability. Results indicated no significant group‐level difference in posttest SAA between the experimental and control groups. More importantly, interaction analysis revealed a significant moderating effect of Initial Self‐Assessment Accuracy (ISAA). Students with lower baseline accuracy benefited substantially from LLMF, while those with higher baseline SAA showed limited change. This compensatory effect highlights the potential of LLMF to reduce inequities in self‐regulated learning. These findings demonstrate the potential of AI‐driven feedback systems to cost‐effectively reduce calibration gaps and foster metacognitive development. By embedding adaptive and personalised mechanisms, such systems advance educational equity and promote scalable personalised learning. They also contribute to the broader agenda of intelligent and sustainable education.
- 7 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 12 (6) , 5375-5388
Early Classification of Time Series Based on Reliable Labels
Early classification of time series aims to predict class labels as promptly as possible while maintaining sufficient accuracy. Existing approaches often rely on multiple classifiers for different input lengths, increasing training cost and complexity. We propose a unified framework comprising: 1) a subsequence classification model for variable-length inputs; 2) a subsequence information enough model; 3) a reliable label module; and 4) a confidence estimation mechanism. Our method dynamically determines when sufficient information is available for reliable classification. Experiments on 45 UCR datasets demonstrate superior performance over baseline methods, especially on SIMULATED, MOTION, and SENSOR tasks.
- 7 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 12 (6) , 5375-5388
Early Classification of Time Series Based on Reliable Labels
Early classification of time series aims to predict class labels as promptly as possible while maintaining sufficient accuracy. Existing approaches often rely on multiple classifiers for different input lengths, increasing training cost and complexity. We propose a unified framework comprising: 1) a subsequence classification model for variable-length inputs; 2) a subsequence information enough model; 3) a reliable label module; and 4) a confidence estimation mechanism. Our method dynamically determines when sufficient information is available for reliable classification. Experiments on 45 UCR datasets demonstrate superior performance over baseline methods, especially on SIMULATED, MOTION, and SENSOR tasks.
- Open Access
- 11 March 2024
- Published by MDPI AG in Future Internet
- Vol. 16 (3) , 95
Personalized Federated Learning with Adaptive Feature Extraction and Category Prediction in Non-IID Datasets
Federated learning trains a neural network model using the client’s data to maintain the benefits of centralized model training while maintaining their privacy. However, if the client data are not independently and identically distributed (non-IID) because of different environments, the accuracy of the model may suffer from client drift during training owing to discrepancies in each client’s data. This study proposes a personalized federated learning algorithm based on the concept of multitask learning to divide each client model into two layers: a feature extraction layer and a category prediction layer. The feature extraction layer maps the input data to a low-dimensional feature vector space. Furthermore, the parameters of the neural network are aggregated with those of other clients using an adaptive method. The category prediction layer maps low-dimensional feature vectors to the label sample space, with its parameters remaining unaffected by other clients to maintain client uniqueness. The proposed personalized federated learning method produces faster learning model convergence rates and higher accuracy rates for the non-IID datasets in our experiments.
- 11 March 2024
- Published by MDPI AG in Future Internet
- Vol. 16 (3) , 95
Open AccessPersonalized Federated Learning with Adaptive Feature Extraction and Category Prediction in Non-IID Datasets
Federated learning trains a neural network model using the client’s data to maintain the benefits of centralized model training while maintaining their privacy. However, if the client data are not independently and identically distributed (non-IID) because of different environments, the accuracy of the model may suffer from client drift during training owing to discrepancies in each client’s data. This study proposes a personalized federated learning algorithm based on the concept of multitask learning to divide each client model into two layers: a feature extraction layer and a category prediction layer. The feature extraction layer maps the input data to a low-dimensional feature vector space. Furthermore, the parameters of the neural network are aggregated with those of other clients using an adaptive method. The category prediction layer maps low-dimensional feature vectors to the label sample space, with its parameters remaining unaffected by other clients to maintain client uniqueness. The proposed personalized federated learning method produces faster learning model convergence rates and higher accuracy rates for the non-IID datasets in our experiments.
- Open Access
- 24 January 2024
- Published by MDPI AG in Preprints.org
Personalized Federated Learning with Adaptive Feature Extraction and Category Prediction in non-IID Datasets
Federated learning trains a neural network model using the client's data to maintain the benefits of centralized model training, while maintaining the privacy. However, if the client data are not independently and identically distributed (non-IID) because of different environments, the accuracy of the model may suffer from client drift during training owing to discrepancies in each client's data. This study proposes a personalized federated learning algorithm based on the concept of multitask learning to divide each client model into two layers: a feature extraction layer and a category prediction layer. The feature extraction layer maps the input data to a low-dimensional feature vector space. Furthermore, the parameters of the neural network are aggregated with those of other clients using an adaptive method. The category prediction layer maps low-dimensional feature vectors to the label sample space, with its parameters remaining unaffected by other clients to maintain client uniqueness. The proposed personalized federated learning method produces faster learning model convergence rates and higher accuracy rates for the non-IID datasets in our experiments.
- 24 January 2024
- Published by MDPI AG in Preprints.org
Open AccessPersonalized Federated Learning with Adaptive Feature Extraction and Category Prediction in non-IID Datasets
Federated learning trains a neural network model using the client's data to maintain the benefits of centralized model training, while maintaining the privacy. However, if the client data are not independently and identically distributed (non-IID) because of different environments, the accuracy of the model may suffer from client drift during training owing to discrepancies in each client's data. This study proposes a personalized federated learning algorithm based on the concept of multitask learning to divide each client model into two layers: a feature extraction layer and a category prediction layer. The feature extraction layer maps the input data to a low-dimensional feature vector space. Furthermore, the parameters of the neural network are aggregated with those of other clients using an adaptive method. The category prediction layer maps low-dimensional feature vectors to the label sample space, with its parameters remaining unaffected by other clients to maintain client uniqueness. The proposed personalized federated learning method produces faster learning model convergence rates and higher accuracy rates for the non-IID datasets in our experiments.