Yin-Tzer Shih
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yin-Tzer Shih has a total of 65 co-authors. Here are the top co-authors by number of shared publications:
Sana Ullah Saqib
4 shared publicationsC.-S. Chien
Affiliation Dept. of Applied Mathematics, National Chung Hsing University, Taichung, Taiwan
3 shared publicationsY.-H. Chen
Department of Applied Mathematics, National Chung Hsing University, Taichung, Taiwan
3 shared publicationsC.-S. Tsai
Affiliation Dept. of Management of Information Systems, National Chung Hsing University, Taichung, Taiwan
2 shared publicationsChih-Ching Tsai
Department of Applied Mathematics, National Chung Hsing University, Taichung, Taiwan
2 shared publicationsHui-Ching Wang
Department of Applied Mathematics, National Chung Hsing University, Taichung, Taiwan; Department of Applied Mathematics, National Chung Hsing University, Taichung, Taiwan
2 shared publications
Latest Publications
- Open Access
Branch-Wise Regularization for Heterogeneous GNN-Based MALDI-TOF AMR Prediction with Biomarker-Consensus Edges
This paper predicts antimicrobial resistance (AMR) from matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectra using DRIAMS, evaluated across 13 species-antibiotic datasets spanning sensitive/resistant imbalance ratios of 2.95:1 to 141.21:1 and raw spectral similarities of 0.9166–0.9946. We propose HetSAGE, a heterogeneous graph neural network (GNN) that combines full-spectrum “raw” edges with consensus “biomarker” edges derived from a 4-method feature-selection vote and applies per-edge-type dropout to regularize the two views separately. HetSAGE outperforms a plain multilayer perceptron (MLP) only when spectral similarity is high, consistent with oversmoothing effects reported in the broader GNN literature; the biomarker view converges on features close to independently established clinical markers: within ±1 Da in Staphylococcus aureus , Oxacillin, and within range in Klebsiella pneumoniae , Meropenem, confirming it is not a black box. At the same time, per-edge-type dropout protects this smaller signal from dilution by regularization tuned for the noisier raw spectrum. Together, these results indicate that heterogeneous graph structure is not a universal win for MALDI-TOF AMR prediction but a conditional one, tied to a measurable property of the data, spectral similarity, rather than to architecture alone. This caveat is largely unaddressed in current DRIAMS-based GNN work and provides a concrete signal for practitioners deciding when a graph model is worth the added complexity over a plain MLP. Author Summary: Diagnosing which antibiotics will work against a bacterial infection typically takes days because it requires growing the bacteria in the lab. A faster technology, MALDI-TOF mass spectrometry, produces a chemical fingerprint of the bacteria within hours, and prior work has shown that this fingerprint alone can predict antibiotic resistance using machine learning. We build on this by connecting samples with similar fingerprints into a graph and using a graph neural network to share information across similar cases. We add a second, more targeted connection type based on a small set of chemical features that a panel of methods agrees are diagnostic, and show that these features align with markers clinicians already use. We find that this graph-based approach helps most when the raw fingerprints are very similar, but offers no advantage otherwise, an important caveat for others building similar systems, since the benefit depends on a measurable property of the data rather than being universal.
Open AccessBranch-Wise Regularization for Heterogeneous GNN-Based MALDI-TOF AMR Prediction with Biomarker-Consensus Edges
This paper predicts antimicrobial resistance (AMR) from matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectra using DRIAMS, evaluated across 13 species-antibiotic datasets spanning sensitive/resistant imbalance ratios of 2.95:1 to 141.21:1 and raw spectral similarities of 0.9166–0.9946. We propose HetSAGE, a heterogeneous graph neural network (GNN) that combines full-spectrum “raw” edges with consensus “biomarker” edges derived from a 4-method feature-selection vote and applies per-edge-type dropout to regularize the two views separately. HetSAGE outperforms a plain multilayer perceptron (MLP) only when spectral similarity is high, consistent with oversmoothing effects reported in the broader GNN literature; the biomarker view converges on features close to independently established clinical markers: within ±1 Da in Staphylococcus aureus , Oxacillin, and within range in Klebsiella pneumoniae , Meropenem, confirming it is not a black box. At the same time, per-edge-type dropout protects this smaller signal from dilution by regularization tuned for the noisier raw spectrum. Together, these results indicate that heterogeneous graph structure is not a universal win for MALDI-TOF AMR prediction but a conditional one, tied to a measurable property of the data, spectral similarity, rather than to architecture alone. This caveat is largely unaddressed in current DRIAMS-based GNN work and provides a concrete signal for practitioners deciding when a graph model is worth the added complexity over a plain MLP. Author Summary: Diagnosing which antibiotics will work against a bacterial infection typically takes days because it requires growing the bacteria in the lab. A faster technology, MALDI-TOF mass spectrometry, produces a chemical fingerprint of the bacteria within hours, and prior work has shown that this fingerprint alone can predict antibiotic resistance using machine learning. We build on this by connecting samples with similar fingerprints into a graph and using a graph neural network to share information across similar cases. We add a second, more targeted connection type based on a small set of chemical features that a panel of methods agrees are diagnostic, and show that these features align with markers clinicians already use. We find that this graph-based approach helps most when the raw fingerprints are very similar, but offers no advantage otherwise, an important caveat for others building similar systems, since the benefit depends on a measurable property of the data rather than being universal.
- 8 September 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 183
A novel hybrid neuro-stochastic framework for human immunodeficiency virus and Hepatitis B Virus co-infection dynamics with vaccination: Computational aspects and dynamical analysis, with artificial neural network-based solutions
- 8 September 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 183
A novel hybrid neuro-stochastic framework for human immunodeficiency virus and Hepatitis B Virus co-infection dynamics with vaccination: Computational aspects and dynamical analysis, with artificial neural network-based solutions
- Open Access
- 8 August 2026
- Published by Springer Nature in International Journal of Dynamics and Control
- Vol. 14 (9) , 324
Computational framework for hearing loss due to mumps virus model using Gudermannian neural networks and optimization techniques
- 8 August 2026
- Published by Springer Nature in International Journal of Dynamics and Control
- Vol. 14 (9) , 324
Open AccessComputational framework for hearing loss due to mumps virus model using Gudermannian neural networks and optimization techniques
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 1674-1685
Transformer Based Encoder Representation for Audio Anti-Spoofing: Separability and Alignment
This paper proposes a Modified Audio Anti-Spoofing method based on a pretrained transformer and the strategy of separability and alignment, which is called MAAS-SA hereafter. Traditional schemes suffer from a challenge on generalizations for cross multiple languages because their models are solely trained by a single language dataset. The MAAS-SA method utilizes three distinct language datasets to overcome the challenge mentioned above. However, a difficulty arises in effectively extracting common features among these three datasets due to the existence of high variations among them. Therefore, the MAAS-SA method comprises three main components to tackle the difficulty, which include a pretrained transformer as an encoder, a combination of the supervised contrastive, hinge, and focal losses to enhance the feature separability, and an alignment classifier with multiple boundaries based on domain generalization to promote classification ability. Moreover, the alignment classifier of the MAAS-SA method not only extracts common features among these three datasets but also identifies unique features for each dataset. Additionally, a variant of the MAAS-SA method, which is called WMAAS-SA, is developed by integrating wavelet domain signals and Kullback-Leibler divergence (KLD) loss to extract more distinguishable representations and then to improve classification ability on audio anti-spoofing. Finally, experimental results demonstrate that the WMAAS-SA method outperforms existing approaches across datasets, ASVspoof 2019 logical access, Chinese Fake Audio Detection (CFAD), French and Russian language in Multi-Language Audio Anti-Spoofing Dataset (MLAAD). Moreover, it is superior to other schemes on unseen attacks and languages.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 1674-1685
Transformer Based Encoder Representation for Audio Anti-Spoofing: Separability and Alignment
This paper proposes a Modified Audio Anti-Spoofing method based on a pretrained transformer and the strategy of separability and alignment, which is called MAAS-SA hereafter. Traditional schemes suffer from a challenge on generalizations for cross multiple languages because their models are solely trained by a single language dataset. The MAAS-SA method utilizes three distinct language datasets to overcome the challenge mentioned above. However, a difficulty arises in effectively extracting common features among these three datasets due to the existence of high variations among them. Therefore, the MAAS-SA method comprises three main components to tackle the difficulty, which include a pretrained transformer as an encoder, a combination of the supervised contrastive, hinge, and focal losses to enhance the feature separability, and an alignment classifier with multiple boundaries based on domain generalization to promote classification ability. Moreover, the alignment classifier of the MAAS-SA method not only extracts common features among these three datasets but also identifies unique features for each dataset. Additionally, a variant of the MAAS-SA method, which is called WMAAS-SA, is developed by integrating wavelet domain signals and Kullback-Leibler divergence (KLD) loss to extract more distinguishable representations and then to improve classification ability on audio anti-spoofing. Finally, experimental results demonstrate that the WMAAS-SA method outperforms existing approaches across datasets, ASVspoof 2019 logical access, Chinese Fake Audio Detection (CFAD), French and Russian language in Multi-Language Audio Anti-Spoofing Dataset (MLAAD). Moreover, it is superior to other schemes on unseen attacks and languages.
- Open Access
- 5 January 2026
- Published by Elsevier in Thermal Science and Engineering Progress
- Vol. 70
Intelligent computational modeling of Maxwell–Smoluchowski thermal slip effects under non-uniform heat sources using radial basis networks
- 5 January 2026
- Published by Elsevier in Thermal Science and Engineering Progress
- Vol. 70
Open AccessIntelligent computational modeling of Maxwell–Smoluchowski thermal slip effects under non-uniform heat sources using radial basis networks
- Open Access
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in AIMS Mathematics
- Vol. 11 (7) , 22688-22723
A data-driven mathematical modeling framework for predicting HIV/HBV coinfection dynamics using machine learning approaches
Viral coinfections remain a significant global health challenge, particularly when infections like human immunodeficiency virus (HIV) coexist with other diseases, causing coinfection in a single individual. This study develops a fractional-order multi-stage HIV/HBV coinfection model using the Caputo fractional derivative to capture memory effects in disease transmission and progression. The model includes latent, asymptomatic, and symptomatic infection stages, as well as key epidemiological factors, including the effects of HBV vaccination, recovery processes among asymptomatic and fully-infected HBV individuals, and the acquisition of secondary infections by exposed, asymptomatic, and individuals. The qualitative properties of the model, such as positivity, boundedness, and sensitivity of key parameters, are analyzed to evaluate the biological feasibility of the system. The basic reproduction numbers for HIV and HBV are calculated, and a sensitivity analysis is conducted to determine the primary parameters affecting coinfection transmission. The coinfected HIV/HBV model is numerically solved using an artificial neural network optimized by the Levenberg-Marquardt algorithm (ANN-LMA). The performance of the ANN-LMA scheme is evaluated using mean squared error (MSE), root mean square error (RMSE), regression analysis, training-state analysis, error histograms, and comparison of overlap results with reference numerical results, demonstrating minimal absolute error. The results underscore the significance of the proposed network in predicting complex coinfection dynamics. This study provides policymakers and epidemic managers with a reliable modeling tool for developing adaptable and sustainable strategies to control HIV/HBV coinfection transmission.
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in AIMS Mathematics
- Vol. 11 (7) , 22688-22723
Open AccessA data-driven mathematical modeling framework for predicting HIV/HBV coinfection dynamics using machine learning approaches
Viral coinfections remain a significant global health challenge, particularly when infections like human immunodeficiency virus (HIV) coexist with other diseases, causing coinfection in a single individual. This study develops a fractional-order multi-stage HIV/HBV coinfection model using the Caputo fractional derivative to capture memory effects in disease transmission and progression. The model includes latent, asymptomatic, and symptomatic infection stages, as well as key epidemiological factors, including the effects of HBV vaccination, recovery processes among asymptomatic and fully-infected HBV individuals, and the acquisition of secondary infections by exposed, asymptomatic, and individuals. The qualitative properties of the model, such as positivity, boundedness, and sensitivity of key parameters, are analyzed to evaluate the biological feasibility of the system. The basic reproduction numbers for HIV and HBV are calculated, and a sensitivity analysis is conducted to determine the primary parameters affecting coinfection transmission. The coinfected HIV/HBV model is numerically solved using an artificial neural network optimized by the Levenberg-Marquardt algorithm (ANN-LMA). The performance of the ANN-LMA scheme is evaluated using mean squared error (MSE), root mean square error (RMSE), regression analysis, training-state analysis, error histograms, and comparison of overlap results with reference numerical results, demonstrating minimal absolute error. The results underscore the significance of the proposed network in predicting complex coinfection dynamics. This study provides policymakers and epidemic managers with a reliable modeling tool for developing adaptable and sustainable strategies to control HIV/HBV coinfection transmission.
- Open Access
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in AIMS Mathematics
- Vol. 11 (1) , 881-906
Application of Tri-layered RNN scheme for Maxwell model subject to MHD
This study examines the characteristics of the Maxwell model under the influence of magnetohydrodynamics (MHD). Due to its inherent viscosity and elasticity, this model has significant applications in both industrial and biological contexts. The core innovation of this work lies in the development and application of a soft computational approach, specifically, the design and implementation of layered supervised recurrent neural networks optimized via the Levenberg-Marquardt (LSRNNs-LMO) technique, to predict the thermodynamic properties of the Maxwell model over a sheet, with particular focus on melting heat and zero mass flux boundary conditions, as inspired by the Cattaneo-Christov heat flux formulation and Lorentz force effects (TMS-MHZCL) model. The LSRNNs-LMO model is trained using data generated through a reliable numerical scheme. Simulation outcomes from the proposed LSRNNs-LMO method show excellent agreement with numerical results across multiple test cases, exhibiting minimal errors and high robustness. The accuracy of the proposed technique is thoroughly validated using error histograms, optimization control curves (mean squared error), root-mean-square error, autocorrelation analysis, regression evaluation, and the Nash-Sutcliffe efficiency (NSE) metric for the TMS-MHZCL model. These assessments provide strong evidence of the predictive validity and precision of the developed LSRNNs-LMO approach.
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in AIMS Mathematics
- Vol. 11 (1) , 881-906
Open AccessApplication of Tri-layered RNN scheme for Maxwell model subject to MHD
This study examines the characteristics of the Maxwell model under the influence of magnetohydrodynamics (MHD). Due to its inherent viscosity and elasticity, this model has significant applications in both industrial and biological contexts. The core innovation of this work lies in the development and application of a soft computational approach, specifically, the design and implementation of layered supervised recurrent neural networks optimized via the Levenberg-Marquardt (LSRNNs-LMO) technique, to predict the thermodynamic properties of the Maxwell model over a sheet, with particular focus on melting heat and zero mass flux boundary conditions, as inspired by the Cattaneo-Christov heat flux formulation and Lorentz force effects (TMS-MHZCL) model. The LSRNNs-LMO model is trained using data generated through a reliable numerical scheme. Simulation outcomes from the proposed LSRNNs-LMO method show excellent agreement with numerical results across multiple test cases, exhibiting minimal errors and high robustness. The accuracy of the proposed technique is thoroughly validated using error histograms, optimization control curves (mean squared error), root-mean-square error, autocorrelation analysis, regression evaluation, and the Nash-Sutcliffe efficiency (NSE) metric for the TMS-MHZCL model. These assessments provide strong evidence of the predictive validity and precision of the developed LSRNNs-LMO approach.
- Open Access
- 1 November 2025
- Published by Elsevier in Journal of Dental Sciences
- Vol. 21 (1) , 457-467
Nuclear factor erythroid 2-related factor 2/heme oxygenase-1 activation by nattokinase reduces pro-inflammatory and matrix-degrading mediators in human gingival fibroblasts
- 1 November 2025
- Published by Elsevier in Journal of Dental Sciences
- Vol. 21 (1) , 457-467
Open AccessNuclear factor erythroid 2-related factor 2/heme oxygenase-1 activation by nattokinase reduces pro-inflammatory and matrix-degrading mediators in human gingival fibroblasts
- Open Access
- 1 September 2025
- Published by SAGE Publications in Journal of Central Nervous System Disease
- Vol. 17
Impact of the COVID-19 Pandemic on 30-Day Readmission Rates in Hospitalized Stroke Patients: A Retrospective Cohort Study
Background: Stroke is one of the leading causes of disability and mortality worldwide, and the 30-day readmission rate is a crucial indicator for assessing the quality of healthcare and patient outcomes. Previous studies have reported a readmission rate of approximately 10%-20%. Objectives: This study investigated the impact of the COVID-19 pandemic on the readmission rate at 30 days among hospitalized stroke patients. Methods: This retrospective cohort study was conducted in a 1500-bed medical center in central Taiwan between January 1, 2020, and December 31, 2023. Data, including 5095 hospitalized stroke patients, 505 readmissions within 30 days, and multiple variables that influence hospital readmission at 30 days, including demographics, clinical characteristics, physiological data, and medication use before discharge, were obtained from hospital electronic health records (EHR). Results: The results revealed a significant decline in 30-day readmission rates between the early (2020∼2021) and latter (2022∼2023) phases of the pandemic, decreasing from 10.71% to 8.97% ( P = .039). This reduction may be attributed to improvements in medical strategies, increased adoption of telemedicine, and improved post-discharge care. Generalized linear regression analysis indicated that male patients were at higher risk of readmission than female patients (OR = 1.28, P = .01), while other variables, such as age, BMI, and blood pressure, did not reach statistical significance. Furthermore, common cardiovascular and metabolic medications (antithrombotic agents, lipid-lowering drugs, antihypertensive drugs, and antidiabetic medications) did not significantly affect the risk of readmission ( P > .05). Conclusions: This finding suggests that improvements in medical management during the early pandemic phase may have contributed to a reduced risk of readmission. Furthermore, male patients had a higher risk of readmission, highlighting the need for improved care strategies for this group.
- 1 September 2025
- Published by SAGE Publications in Journal of Central Nervous System Disease
- Vol. 17
Open AccessImpact of the COVID-19 Pandemic on 30-Day Readmission Rates in Hospitalized Stroke Patients: A Retrospective Cohort Study
Background: Stroke is one of the leading causes of disability and mortality worldwide, and the 30-day readmission rate is a crucial indicator for assessing the quality of healthcare and patient outcomes. Previous studies have reported a readmission rate of approximately 10%-20%. Objectives: This study investigated the impact of the COVID-19 pandemic on the readmission rate at 30 days among hospitalized stroke patients. Methods: This retrospective cohort study was conducted in a 1500-bed medical center in central Taiwan between January 1, 2020, and December 31, 2023. Data, including 5095 hospitalized stroke patients, 505 readmissions within 30 days, and multiple variables that influence hospital readmission at 30 days, including demographics, clinical characteristics, physiological data, and medication use before discharge, were obtained from hospital electronic health records (EHR). Results: The results revealed a significant decline in 30-day readmission rates between the early (2020∼2021) and latter (2022∼2023) phases of the pandemic, decreasing from 10.71% to 8.97% ( P = .039). This reduction may be attributed to improvements in medical strategies, increased adoption of telemedicine, and improved post-discharge care. Generalized linear regression analysis indicated that male patients were at higher risk of readmission than female patients (OR = 1.28, P = .01), while other variables, such as age, BMI, and blood pressure, did not reach statistical significance. Furthermore, common cardiovascular and metabolic medications (antithrombotic agents, lipid-lowering drugs, antihypertensive drugs, and antidiabetic medications) did not significantly affect the risk of readmission ( P > .05). Conclusions: This finding suggests that improvements in medical management during the early pandemic phase may have contributed to a reduced risk of readmission. Furthermore, male patients had a higher risk of readmission, highlighting the need for improved care strategies for this group.