Ibrahim Karabayir
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
Ibrahim Karabayir has a total of 89 co-authors. Here are the top co-authors by number of shared publications:
- Oguz Akbilgic
Department of Cardiovascular Medicine, Wake Forest School of Medicine , 486 Patterson Avenue, Suite #446, Winston-Salem, NC 27157
10 shared publications - Erin D. Myers
College of Medicine, The University of Tennessee Health Science Center, Memphis, USA
4 shared publications - Joseph Francis Pierre
University of Wisconsin-Madison
4 shared publications - Kent A. Willis
The University of Alabama at Birmingham
4 shared publications - Ajay J. Talati
Division of Neonatology, Department of Pediatrics, College of Medicine, The University of Tennessee Health Science Center, Memphis, USA
3 shared publications - Brian M. Peters
Department of Clinical Pharmacy and Translational Science, College of Pharmacy, University of Tennessee Health Science Center
3 shared publications
Latest Publications
- Open Access
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
Electrocardiographic artificial intelligence for prediction of pacemaker implantation after transcatheter aortic valve replacement: a retrospective multi-center study Supplementary Appendix
Background: Permanent pacemaker (PPM) implantation remains a common complication after transcatheter aortic valve replacement (TAVR). Existing clinical and imaging-based risk prediction models showing limited and inconsistent performance. We evaluated whether a pre-procedural ECG-based artificial intelligence (ECG-AI) model predicts 30-day PPM post-TAVR. Methods: In this retrospective multicenter study, a 1-dimensional convolutional neural network was trained on 34,745 ECGs from 5,932 patients with moderate-to-severe aortic stenosis (proxy cohort) to enable prediction of 30-day PPM implantation after TAVR using preprocedural ECGs. Zero-shot external validation was performed in two independent TAVR cohorts (Atrium Health Wake Forest Baptist [AHWFB] and Sanger Heart and Vascular Institute [SHVI]) without recalibration or fine-tuning. Model performance was assessed using discrimination and calibration metrics, with subgroup analyses by valve platform, clinical factors, and ECG-to-TAVR timing. Incremental and independent prognostic value was evaluated using multivariable logistic regression incorporating clinical and imaging covariates. Findings: The model was evaluated in two heterogeneous TAVR cohorts (AHWFB, n=823; SHVI, n=162) with differences in baseline characteristics and valve use, demonstrating moderate but consistent discrimination across cohorts (AHWFB: AUROC 0·67 [0·61–0·73], Brier 0·11 [0·09–0·13]; SHVI: AUROC 0·78 [0·67–0·88], Brier 0·07 [0·04–0·11]) with high negative predictive value (0·92–0·99). Performance remained stable for ECGs obtained within 0–2 days pre-TAVR (AUROC ~0·75–0·78) and in Sapien-only analyses (AHWFB: 0·77 [0·70–0·84]; SHVI: 0·79 [0·68–0·89]). In multivariable models, baseline covariates modestly improved discrimination (ΔAUC ≤0·06 in SHVI), while ECG-AI remained the only independent and consistent predictor (OR 1·36–2·18). Interpretation: An ECG-AI model trained in a large cohort demonstrated robust, transferable prediction of 30-day PPM post-TAVR using pre-procedural ECGs across multiple TAVR cohorts without recalibration. It captured prognostic information beyond traditional baseline risk factors and showed consistent independent and incremental predictive value.
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessElectrocardiographic artificial intelligence for prediction of pacemaker implantation after transcatheter aortic valve replacement: a retrospective multi-center study Supplementary Appendix
Background: Permanent pacemaker (PPM) implantation remains a common complication after transcatheter aortic valve replacement (TAVR). Existing clinical and imaging-based risk prediction models showing limited and inconsistent performance. We evaluated whether a pre-procedural ECG-based artificial intelligence (ECG-AI) model predicts 30-day PPM post-TAVR. Methods: In this retrospective multicenter study, a 1-dimensional convolutional neural network was trained on 34,745 ECGs from 5,932 patients with moderate-to-severe aortic stenosis (proxy cohort) to enable prediction of 30-day PPM implantation after TAVR using preprocedural ECGs. Zero-shot external validation was performed in two independent TAVR cohorts (Atrium Health Wake Forest Baptist [AHWFB] and Sanger Heart and Vascular Institute [SHVI]) without recalibration or fine-tuning. Model performance was assessed using discrimination and calibration metrics, with subgroup analyses by valve platform, clinical factors, and ECG-to-TAVR timing. Incremental and independent prognostic value was evaluated using multivariable logistic regression incorporating clinical and imaging covariates. Findings: The model was evaluated in two heterogeneous TAVR cohorts (AHWFB, n=823; SHVI, n=162) with differences in baseline characteristics and valve use, demonstrating moderate but consistent discrimination across cohorts (AHWFB: AUROC 0·67 [0·61–0·73], Brier 0·11 [0·09–0·13]; SHVI: AUROC 0·78 [0·67–0·88], Brier 0·07 [0·04–0·11]) with high negative predictive value (0·92–0·99). Performance remained stable for ECGs obtained within 0–2 days pre-TAVR (AUROC ~0·75–0·78) and in Sapien-only analyses (AHWFB: 0·77 [0·70–0·84]; SHVI: 0·79 [0·68–0·89]). In multivariable models, baseline covariates modestly improved discrimination (ΔAUC ≤0·06 in SHVI), while ECG-AI remained the only independent and consistent predictor (OR 1·36–2·18). Interpretation: An ECG-AI model trained in a large cohort demonstrated robust, transferable prediction of 30-day PPM post-TAVR using pre-procedural ECGs across multiple TAVR cohorts without recalibration. It captured prognostic information beyond traditional baseline risk factors and showed consistent independent and incremental predictive value.
- Open Access
- 27 January 2025
- Published by Springer Nature in Microbiome
- Vol. 13 (1) , 1-20
The fungal microbiota modulate neonatal oxygen-induced lung injury
The immature lungs of very preterm infants are exposed to supraphysiologic oxygen, contributing to bronchopulmonary dysplasia (BPD), a chronic lung disease that is the most common morbidity of prematurity. While the microbiota significantly influences neonatal health, the relationship between the intestinal microbiome, particularly micro-eukaryotic members such as fungi and yeast, and lung injury severity in newborns remains unknown. Here, we show that the fungal microbiota modulates hyperoxia-induced lung injury severity in very low birth weight premature infants and preclinical pseudohumanized and altered fungal colonization mouse models. Instead of fungal communities dominated by Candida and Saccharomyces, the first stool microbiomes of infants who developed BPD had less interconnected community architectures with a greater diversity of rarer fungi. After using a pseudohumanized model to show that transfer to the neonatal microbiome from infants with BPD increased the severity of lung injury, we used gain and loss of function approaches to demonstrate that modulating the extent of initial neonatal fungal colonization affected the extent of BPD-like lung injury in mice. We also identified alterations in the murine intestinal microbiome and transcriptome associated with augmented lung injury. These findings demonstrate that features of the initial intestinal fungal microbiome are associated with the later development of BPD in premature neonates and exert a microbiome-driven effect that is transferable and modifiable in murine models, which suggests both causality and a potential therapeutic strategy.
- 27 January 2025
- Published by Springer Nature in Microbiome
- Vol. 13 (1) , 1-20
Open AccessThe fungal microbiota modulate neonatal oxygen-induced lung injury
The immature lungs of very preterm infants are exposed to supraphysiologic oxygen, contributing to bronchopulmonary dysplasia (BPD), a chronic lung disease that is the most common morbidity of prematurity. While the microbiota significantly influences neonatal health, the relationship between the intestinal microbiome, particularly micro-eukaryotic members such as fungi and yeast, and lung injury severity in newborns remains unknown. Here, we show that the fungal microbiota modulates hyperoxia-induced lung injury severity in very low birth weight premature infants and preclinical pseudohumanized and altered fungal colonization mouse models. Instead of fungal communities dominated by Candida and Saccharomyces, the first stool microbiomes of infants who developed BPD had less interconnected community architectures with a greater diversity of rarer fungi. After using a pseudohumanized model to show that transfer to the neonatal microbiome from infants with BPD increased the severity of lung injury, we used gain and loss of function approaches to demonstrate that modulating the extent of initial neonatal fungal colonization affected the extent of BPD-like lung injury in mice. We also identified alterations in the murine intestinal microbiome and transcriptome associated with augmented lung injury. These findings demonstrate that features of the initial intestinal fungal microbiome are associated with the later development of BPD in premature neonates and exert a microbiome-driven effect that is transferable and modifiable in murine models, which suggests both causality and a potential therapeutic strategy.
- 4 March 2024
- Published by Elsevier in American Journal of Obstetrics & Gynecology MFM
- Vol. 6 (4) , 101337
Development and validation of an electrocardiographic artificial intelligence model for detection of peripartum cardiomyopathy
- 4 March 2024
- Published by Elsevier in American Journal of Obstetrics & Gynecology MFM
- Vol. 6 (4) , 101337
Development and validation of an electrocardiographic artificial intelligence model for detection of peripartum cardiomyopathy
- Open Access
The fungal intestinal microbiota predict the development of bronchopulmonary dysplasia in very low birthweight newborns
Rationale: Bronchopulmonary dysplasia (BPD) is the most common morbidity affecting very preterm infants. Gut fungal and bacterial microbial communities contribute to multiple lung diseases and may influence BPD pathogenesis. Methods: We performed a prospective, observational cohort study comparing the multikingdom fecal microbiota of 144 preterm infants with or without moderate to severe BPD by sequencing the bacterial 16S and fungal ITS2 ribosomal RNA gene. To address the potential causative relationship between gut dysbiosis and BPD, we used fecal microbiota transplant in an antibiotic-pseudohumanized mouse model. Comparisons were made using RNA sequencing, confocal microscopy, lung morphometry, and oscillometry. Results: We analyzed 102 fecal microbiome samples collected during the second week of life. Infants who later developed BPD showed an obvious fungal dysbiosis as compared to infants without BPD (NoBPD, p = 0.0398, permutational multivariate ANOVA). Instead of fungal communities dominated by Candida and Saccharomyces , the microbiota of infants who developed BPD were characterized by a greater diversity of rarer fungi in less interconnected community architectures. On successful colonization, the gut microbiota from infants with BPD augmented lung injury in the offspring of recipient animals. We identified alterations in the murine intestinal microbiome and transcriptome associated with augmented lung injury. Conclusions: The gut fungal microbiome of infants who will develop BPD is dysbiotic and may contribute to disease pathogenesis.Conclusions: The gut fungal microbiome of infants who will develop BPD is dysbiotic and may contribute to disease pathogenesis.
Open AccessThe fungal intestinal microbiota predict the development of bronchopulmonary dysplasia in very low birthweight newborns
Rationale: Bronchopulmonary dysplasia (BPD) is the most common morbidity affecting very preterm infants. Gut fungal and bacterial microbial communities contribute to multiple lung diseases and may influence BPD pathogenesis. Methods: We performed a prospective, observational cohort study comparing the multikingdom fecal microbiota of 144 preterm infants with or without moderate to severe BPD by sequencing the bacterial 16S and fungal ITS2 ribosomal RNA gene. To address the potential causative relationship between gut dysbiosis and BPD, we used fecal microbiota transplant in an antibiotic-pseudohumanized mouse model. Comparisons were made using RNA sequencing, confocal microscopy, lung morphometry, and oscillometry. Results: We analyzed 102 fecal microbiome samples collected during the second week of life. Infants who later developed BPD showed an obvious fungal dysbiosis as compared to infants without BPD (NoBPD, p = 0.0398, permutational multivariate ANOVA). Instead of fungal communities dominated by Candida and Saccharomyces , the microbiota of infants who developed BPD were characterized by a greater diversity of rarer fungi in less interconnected community architectures. On successful colonization, the gut microbiota from infants with BPD augmented lung injury in the offspring of recipient animals. We identified alterations in the murine intestinal microbiome and transcriptome associated with augmented lung injury. Conclusions: The gut fungal microbiome of infants who will develop BPD is dysbiotic and may contribute to disease pathogenesis.Conclusions: The gut fungal microbiome of infants who will develop BPD is dysbiotic and may contribute to disease pathogenesis.
- Open Access
- 6 July 2022
- Published by Springer Nature in Research Square
Externally Validated AI Model to Identify Prodromal Parkinson’s Disease from ECG
Autonomic nervous system pathology manifests early in Parkinson’s disease (PD) course. Although heart rate variability measured by a 5-minute electrocardiogram (ECG) is reported to be reduced in PD, little is known about ECG markers during prodromal stage, and brief 10-second ECGs have been rarely studied. The aim of the study is to build externally validated ECG based fully automatic artificial intelligence (AI) model to predict PD risk during prodromal stage. We obtained data for PD and date/age/sex/race-matched control subjects from Loyola University Chicago (LUC) and University of Tennessee-Methodist Le Bonheur Healthcare (MLH). We used AI to predict PD risk from standard 10-second 12-lead ECGs performed between 6-months-5 years before diagnosis. The prediction model was built with MLH data and externally validated with LUC data. We identified 131 cases/1058 controls at MLH and 29 cases/165 controls at LUC. We initially trained models on 90% of the MLH data and internally validated them in the remaining 10%. The best performing model with an internal validation was a convolutional neural network model. External validation with LUC data yielded an AUC of 0.67. When we considered only ECGs obtained 6-months to 1-year preceding PD diagnosis, the external validation AUC was 0.74. Using only ECG, we built a predictive model that correctly classified individuals with prodromal PD with moderate accuracy. The model was effective in an independent cohort, particularly closer to disease diagnosis. Standard ECGs may help to identify individuals with prodromal PD for inclusion in disease-modifying therapeutic trials.
- 6 July 2022
- Published by Springer Nature in Research Square
Open AccessExternally Validated AI Model to Identify Prodromal Parkinson’s Disease from ECG
Autonomic nervous system pathology manifests early in Parkinson’s disease (PD) course. Although heart rate variability measured by a 5-minute electrocardiogram (ECG) is reported to be reduced in PD, little is known about ECG markers during prodromal stage, and brief 10-second ECGs have been rarely studied. The aim of the study is to build externally validated ECG based fully automatic artificial intelligence (AI) model to predict PD risk during prodromal stage. We obtained data for PD and date/age/sex/race-matched control subjects from Loyola University Chicago (LUC) and University of Tennessee-Methodist Le Bonheur Healthcare (MLH). We used AI to predict PD risk from standard 10-second 12-lead ECGs performed between 6-months-5 years before diagnosis. The prediction model was built with MLH data and externally validated with LUC data. We identified 131 cases/1058 controls at MLH and 29 cases/165 controls at LUC. We initially trained models on 90% of the MLH data and internally validated them in the remaining 10%. The best performing model with an internal validation was a convolutional neural network model. External validation with LUC data yielded an AUC of 0.67. When we considered only ECGs obtained 6-months to 1-year preceding PD diagnosis, the external validation AUC was 0.74. Using only ECG, we built a predictive model that correctly classified individuals with prodromal PD with moderate accuracy. The model was effective in an independent cohort, particularly closer to disease diagnosis. Standard ECGs may help to identify individuals with prodromal PD for inclusion in disease-modifying therapeutic trials.
- 1 December 2021
- Published by American Society of Clinical Oncology (ASCO) in JCO Clinical Cancer Informatics
- Vol. 5 (5) , 459-468
Artificial Intelligence–Assisted Prediction of Late-Onset Cardiomyopathy Among Childhood Cancer Survivors
PURPOSE: Early identification of childhood cancer survivors at high risk for treatment-related cardiomyopathy may improve outcomes by enabling intervention before development of heart failure. We implemented artificial intelligence (AI) methods using the Children's Oncology Group guideline–recommended baseline ECG to predict cardiomyopathy.MATERIAL AND METHODS: Seven AI and signal processing methods were applied to 10-second 12-lead ECGs obtained on 1,217 adult survivors of childhood cancer prospectively followed in the St Jude Lifetime Cohort (SJLIFE) study. Clinical and echocardiographic assessment of cardiac function was performed at initial and follow-up SJLIFE visits. Cardiomyopathy was defined as an ejection fraction < 50% or an absolute drop from baseline ≥ 10%. Genetic algorithm was used for feature selection, and extreme gradient boosting was applied to predict cardiomyopathy during the follow-up period. Model performance was evaluated by five-fold stratified cross-validation.RESULTS: The median age at baseline SJLIFE evaluation was 31.7 years (range 18.4-66.4), and the time between baseline and follow-up evaluations was 5.2 years (0.5-9.5). Two thirds (67.1%) of patients were exposed to chest radiation, and 76.6% to anthracycline chemotherapy. One hundred seventeen (9.6%) patients developed cardiomyopathy during follow-up. In the model based solely on ECG features, the cross-validation area under the curve (AUC) was 0.87 (95% CI, 0.83 to 0.90), whereas the model based on clinical features had an AUC of 0.69 (95% CI, 0.64 to 0.74). In the model based on ECG and clinical features, the cross-validation AUC was 0.89 (95% CI, 0.86 to 0.91), with a sensitivity of 78% and a specificity of 81%.CONCLUSION: AI using ECG data may assist in the identification of childhood cancer survivors at increased risk for developing future cardiomyopathy.
- 1 December 2021
- Published by American Society of Clinical Oncology (ASCO) in JCO Clinical Cancer Informatics
- Vol. 5 (5) , 459-468
Artificial Intelligence–Assisted Prediction of Late-Onset Cardiomyopathy Among Childhood Cancer Survivors
PURPOSE: Early identification of childhood cancer survivors at high risk for treatment-related cardiomyopathy may improve outcomes by enabling intervention before development of heart failure. We implemented artificial intelligence (AI) methods using the Children's Oncology Group guideline–recommended baseline ECG to predict cardiomyopathy.MATERIAL AND METHODS: Seven AI and signal processing methods were applied to 10-second 12-lead ECGs obtained on 1,217 adult survivors of childhood cancer prospectively followed in the St Jude Lifetime Cohort (SJLIFE) study. Clinical and echocardiographic assessment of cardiac function was performed at initial and follow-up SJLIFE visits. Cardiomyopathy was defined as an ejection fraction < 50% or an absolute drop from baseline ≥ 10%. Genetic algorithm was used for feature selection, and extreme gradient boosting was applied to predict cardiomyopathy during the follow-up period. Model performance was evaluated by five-fold stratified cross-validation.RESULTS: The median age at baseline SJLIFE evaluation was 31.7 years (range 18.4-66.4), and the time between baseline and follow-up evaluations was 5.2 years (0.5-9.5). Two thirds (67.1%) of patients were exposed to chest radiation, and 76.6% to anthracycline chemotherapy. One hundred seventeen (9.6%) patients developed cardiomyopathy during follow-up. In the model based solely on ECG features, the cross-validation area under the curve (AUC) was 0.87 (95% CI, 0.83 to 0.90), whereas the model based on clinical features had an AUC of 0.69 (95% CI, 0.64 to 0.74). In the model based on ECG and clinical features, the cross-validation AUC was 0.89 (95% CI, 0.86 to 0.91), with a sensitivity of 78% and a specificity of 81%.CONCLUSION: AI using ECG data may assist in the identification of childhood cancer survivors at increased risk for developing future cardiomyopathy.
- Open Access
- 9 October 2021
- Published by Oxford University Press (OUP) in European Heart Journal - Digital Health
- Vol. 2 (4) , 626-634
ECG-AI: electrocardiographic artificial intelligence model for prediction of heart failure
Aims: Heart failure (HF) is a leading cause of death. Early intervention is the key to reduce HF-related morbidity and mortality. This study assesses the utility of electrocardiograms (ECGs) in HF risk prediction.Methods and results: Data from the baseline visits (1987–89) of the Atherosclerosis Risk in Communities (ARIC) study was used. Incident hospitalized HF events were ascertained by ICD codes. Participants with good quality baseline ECGs were included. Participants with prevalent HF were excluded. ECG-artificial intelligence (AI) model to predict HF was created as a deep residual convolutional neural network (CNN) utilizing standard 12-lead ECG. The area under the receiver operating characteristic curve (AUC) was used to evaluate prediction models including (CNN), light gradient boosting machines (LGBM), and Cox proportional hazards regression. A total of 14 613 (45% male, 73% of white, mean age ± standard deviation of 54 ± 5) participants were eligible. A total of 803 (5.5%) participants developed HF within 10 years from baseline. Convolutional neural network utilizing solely ECG achieved an AUC of 0.756 (0.717–0.795) on the hold-out test data. ARIC and Framingham Heart Study (FHS) HF risk calculators yielded AUC of 0.802 (0.750–0.850) and 0.780 (0.740–0.830). The highest AUC of 0.818 (0.778–0.859) was obtained when ECG-AI model output, age, gender, race, body mass index, smoking status, prevalent coronary heart disease, diabetes mellitus, systolic blood pressure, and heart rate were used as predictors of HF within LGBM. The ECG-AI model output was the most important predictor of HF.Conclusions: ECG-AI model based solely on information extracted from ECG independently predicts HF with accuracy comparable to existing FHS and ARIC risk calculators.
- 9 October 2021
- Published by Oxford University Press (OUP) in European Heart Journal - Digital Health
- Vol. 2 (4) , 626-634
Open AccessECG-AI: electrocardiographic artificial intelligence model for prediction of heart failure
Aims: Heart failure (HF) is a leading cause of death. Early intervention is the key to reduce HF-related morbidity and mortality. This study assesses the utility of electrocardiograms (ECGs) in HF risk prediction.Methods and results: Data from the baseline visits (1987–89) of the Atherosclerosis Risk in Communities (ARIC) study was used. Incident hospitalized HF events were ascertained by ICD codes. Participants with good quality baseline ECGs were included. Participants with prevalent HF were excluded. ECG-artificial intelligence (AI) model to predict HF was created as a deep residual convolutional neural network (CNN) utilizing standard 12-lead ECG. The area under the receiver operating characteristic curve (AUC) was used to evaluate prediction models including (CNN), light gradient boosting machines (LGBM), and Cox proportional hazards regression. A total of 14 613 (45% male, 73% of white, mean age ± standard deviation of 54 ± 5) participants were eligible. A total of 803 (5.5%) participants developed HF within 10 years from baseline. Convolutional neural network utilizing solely ECG achieved an AUC of 0.756 (0.717–0.795) on the hold-out test data. ARIC and Framingham Heart Study (FHS) HF risk calculators yielded AUC of 0.802 (0.750–0.850) and 0.780 (0.740–0.830). The highest AUC of 0.818 (0.778–0.859) was obtained when ECG-AI model output, age, gender, race, body mass index, smoking status, prevalent coronary heart disease, diabetes mellitus, systolic blood pressure, and heart rate were used as predictors of HF within LGBM. The ECG-AI model output was the most important predictor of HF.Conclusions: ECG-AI model based solely on information extracted from ECG independently predicts HF with accuracy comparable to existing FHS and ARIC risk calculators.
- 2 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 32 (2) , 685-694
A Novel Learning Algorithm to Optimize Deep Neural Networks: Evolved Gradient Direction Optimizer (EVGO)
Gradient-based algorithms have been widely used in optimizing parameters of deep neural networks' (DNNs) architectures. However, the vanishing gradient remains as one of the common issues in the parameter optimization of such networks. To cope with the vanishing gradient problem, in this article, we propose a novel algorithm, evolved gradient direction optimizer (EVGO), updating the weights of DNNs based on the first-order gradient and a novel hyperplane we introduce. We compare the EVGO algorithm with other gradient-based algorithms, such as gradient descent, RMSProp, Adagrad, momentum, and Adam on the well-known Modified National Institute of Standards and Technology (MNIST) data set for handwritten digit recognition by implementing deep convolutional neural networks. Furthermore, we present empirical evaluations of EVGO on the CIFAR-10 and CIFAR-100 data sets by using the well-known AlexNet and ResNet architectures. Finally, we implement an empirical analysis for EVGO and other algorithms to investigate the behavior of the loss functions. The results show that EVGO outperforms all the algorithms in comparison for all experiments. We conclude that EVGO can be used effectively in the optimization of DNNs, and also, the proposed hyperplane may provide a basis for future optimization algorithms.
- 2 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 32 (2) , 685-694
A Novel Learning Algorithm to Optimize Deep Neural Networks: Evolved Gradient Direction Optimizer (EVGO)
Gradient-based algorithms have been widely used in optimizing parameters of deep neural networks' (DNNs) architectures. However, the vanishing gradient remains as one of the common issues in the parameter optimization of such networks. To cope with the vanishing gradient problem, in this article, we propose a novel algorithm, evolved gradient direction optimizer (EVGO), updating the weights of DNNs based on the first-order gradient and a novel hyperplane we introduce. We compare the EVGO algorithm with other gradient-based algorithms, such as gradient descent, RMSProp, Adagrad, momentum, and Adam on the well-known Modified National Institute of Standards and Technology (MNIST) data set for handwritten digit recognition by implementing deep convolutional neural networks. Furthermore, we present empirical evaluations of EVGO on the CIFAR-10 and CIFAR-100 data sets by using the well-known AlexNet and ResNet architectures. Finally, we implement an empirical analysis for EVGO and other algorithms to investigate the behavior of the loss functions. The results show that EVGO outperforms all the algorithms in comparison for all experiments. We conclude that EVGO can be used effectively in the optimization of DNNs, and also, the proposed hyperplane may provide a basis for future optimization algorithms.
- Open Access
- 1 September 2019
- Published by Elsevier in Kidney International Reports
- Vol. 4 (9) , 1219-1229
Machine Learning to Identify Dialysis Patients at High Death Risk
- 1 September 2019
- Published by Elsevier in Kidney International Reports
- Vol. 4 (9) , 1219-1229
Open AccessMachine Learning to Identify Dialysis Patients at High Death Risk