Oğuz Altun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Oğuz Altun has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Ayşe Kale
Beykent University
2 shared publications - Bahadır Uçan
Yildiz Teknik Üniversitesi
1 shared publication - Batuhan Hangun
Department of Computer Engineering, Yildiz Technical University, Istanbul, Turkey
1 shared publication - Betül Akyol
Faculty of Sport Sciences, İnönü University, Malatya, Türkiye
1 shared publication - Betül Aydoğdu
YILDIZ TEKNİK ÜNİVERSİTESİ, SANAT VE TASARIM FAKÜLTESİ, SANAT BÖLÜMÜ, GRAFİK TASARIMI PR
1 shared publication - Elif Çalişkan
Yildiz Teknik Üniversitesi
1 shared publication
Latest Publications
- Open Access
- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
Benchmarking quantum neural networks for wind power prediction: Generalization stability and noise robustness under data-scarce conditions
Accurate wind power prediction is essential for reliable grid integration, yet classical machine learning models often saturate in accuracy and generalize unstably in data-constrained settings such as newly commissioned wind farms. This work benchmarks a 16-parameter quantum neural network (QNN) against six classical regressors, namely support vector regression (SVR), decision tree regression (DTR), random forest (RF), extreme gradient boosting (XGBoost), and two artificial neural networks (ANNs), on real turbine data across four regression metrics, ten random seeds, and four training-set sizes, with significance assessed by Wilcoxon signed-rank tests. Classical ensembles achieve the highest peak accuracy ($R^2 \approx 0.949$) but exhibit diminishing returns and large generalization gaps ($\Delta R^2$ up to $0.034 \pm 0.017$ for RF), indicating overfitting. The QNN is stable across all seeds ($R^2 = 0.902 \pm 0.016$), whereas a parameter-matched ANN with 19 parameters is catastrophically unstable ($\sigma = 0.294$). The QNN also attains one of the smallest generalization gaps ($\Delta R^2 = 0.003 \pm 0.020$), statistically indistinguishable from that of a regularized production-grade ANN with several thousand parameters and significantly smaller than those of the tree-based and boosting baselines ($p < 0.01$). Under depolarizing gate noise the QNN retains $R^2 > 0.89$ at near-term realistic gate error rates ($p = 0.001$), degrading by only 1.1\%, which defines an operational feasibility envelope ($p \leq 0.005$) for near-term quantum hardware. A two-parameter output calibration requiring no retraining restores $R^2 \approx 0.91$ up to $p = 0.02$, outperforming noise-aware retraining and extending the practical envelope fourfold. Classical ensembles remain more accurate, but the QNN offers a stable, low-variance alternative for data-scarce deployment.- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessBenchmarking quantum neural networks for wind power prediction: Generalization stability and noise robustness under data-scarce conditions
Accurate wind power prediction is essential for reliable grid integration, yet classical machine learning models often saturate in accuracy and generalize unstably in data-constrained settings such as newly commissioned wind farms. This work benchmarks a 16-parameter quantum neural network (QNN) against six classical regressors, namely support vector regression (SVR), decision tree regression (DTR), random forest (RF), extreme gradient boosting (XGBoost), and two artificial neural networks (ANNs), on real turbine data across four regression metrics, ten random seeds, and four training-set sizes, with significance assessed by Wilcoxon signed-rank tests. Classical ensembles achieve the highest peak accuracy ($R^2 \approx 0.949$) but exhibit diminishing returns and large generalization gaps ($\Delta R^2$ up to $0.034 \pm 0.017$ for RF), indicating overfitting. The QNN is stable across all seeds ($R^2 = 0.902 \pm 0.016$), whereas a parameter-matched ANN with 19 parameters is catastrophically unstable ($\sigma = 0.294$). The QNN also attains one of the smallest generalization gaps ($\Delta R^2 = 0.003 \pm 0.020$), statistically indistinguishable from that of a regularized production-grade ANN with several thousand parameters and significantly smaller than those of the tree-based and boosting baselines ($p < 0.01$). Under depolarizing gate noise the QNN retains $R^2 > 0.89$ at near-term realistic gate error rates ($p = 0.001$), degrading by only 1.1\%, which defines an operational feasibility envelope ($p \leq 0.005$) for near-term quantum hardware. A two-parameter output calibration requiring no retraining restores $R^2 \approx 0.91$ up to $p = 0.02$, outperforming noise-aware retraining and extending the practical envelope fourfold. Classical ensembles remain more accurate, but the QNN offers a stable, low-variance alternative for data-scarce deployment. - 13 September 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 138
An efficient identity-preserving and fast-converging hybrid generative adversarial network inversion framework
- 13 September 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 138
An efficient identity-preserving and fast-converging hybrid generative adversarial network inversion framework
- 31 July 2023
- Published by Uludag Universitesi Fen Edebiyat Fakultesi Sosyal Bilimler Dergisi in Uludağ Üniversitesi Fen-Edebiyat Fakültesi Sosyal Bilimler Dergisi
- Vol. 24 (45) , 539-561
EĞİTİM ODAKLI SANAL EVREN TASARIMI: YILDIZ TEKNİK ÜNİVERSİTESİ ÖRNEĞİ
Dijital teknolojilerin eğitim dünyasında etkisini hızla artırması sonucunda sanal gerçeklik teknolojisinin eğitimde kullanımını görmekteyiz. Çok çeşitli etkileşimler barındıran ve fiziksel olarak bir mekâna bağlı kalmadan uzaktan yaşanabilecek deneyimler sunan metaverse ortamları, üniversitede işlenecek teorik ve uygulamalı eğitimler konusunda da şüphesiz büyük önem taşımaktadır. Bu çalışmada, kullanıcı-odaklı tasarım anlayışla eğitim odaklı sanal evren (metaverse) tasarım sürecinde yeni bir tasarım yaklaşımı ortaya koymaktır. Bu bağlamda, Yıldız Teknik Üniversitesi için bir sanal sınıf tasarımı örnek olarak sunulmaktadır. Durum çalışması olarak gerçekleştirilen çalışmada, ilgili alanyazın araştırılmış, üniversite öğrencilerinin görüşleri alınmış ve eğitim amaçlı mevcut sanal gerçeklik ortamları incelenmiştir. Toplamda 130 öğrencinin görüşleri, açık uçlu sorulardan oluşan bir anket yoluyla alınmış ve içerik analizi yöntemiyle analiz edilmiştir. Analizler sonucunda, öğrencilerin metaverse ortamında bir eğitim platformuna yönelik ihtiyaç ve talepleri 11 temadan oluşan 3 kategoride toplanarak değerlendirilmiştir. Analiz sonuçlarına göre metaverse ortamında bir eğitim platformu ve sınıf ortamı tasarlanmıştır.
- 31 July 2023
- Published by Uludag Universitesi Fen Edebiyat Fakultesi Sosyal Bilimler Dergisi in Uludağ Üniversitesi Fen-Edebiyat Fakültesi Sosyal Bilimler Dergisi
- Vol. 24 (45) , 539-561
EĞİTİM ODAKLI SANAL EVREN TASARIMI: YILDIZ TEKNİK ÜNİVERSİTESİ ÖRNEĞİ
Dijital teknolojilerin eğitim dünyasında etkisini hızla artırması sonucunda sanal gerçeklik teknolojisinin eğitimde kullanımını görmekteyiz. Çok çeşitli etkileşimler barındıran ve fiziksel olarak bir mekâna bağlı kalmadan uzaktan yaşanabilecek deneyimler sunan metaverse ortamları, üniversitede işlenecek teorik ve uygulamalı eğitimler konusunda da şüphesiz büyük önem taşımaktadır. Bu çalışmada, kullanıcı-odaklı tasarım anlayışla eğitim odaklı sanal evren (metaverse) tasarım sürecinde yeni bir tasarım yaklaşımı ortaya koymaktır. Bu bağlamda, Yıldız Teknik Üniversitesi için bir sanal sınıf tasarımı örnek olarak sunulmaktadır. Durum çalışması olarak gerçekleştirilen çalışmada, ilgili alanyazın araştırılmış, üniversite öğrencilerinin görüşleri alınmış ve eğitim amaçlı mevcut sanal gerçeklik ortamları incelenmiştir. Toplamda 130 öğrencinin görüşleri, açık uçlu sorulardan oluşan bir anket yoluyla alınmış ve içerik analizi yöntemiyle analiz edilmiştir. Analizler sonucunda, öğrencilerin metaverse ortamında bir eğitim platformuna yönelik ihtiyaç ve talepleri 11 temadan oluşan 3 kategoride toplanarak değerlendirilmiştir. Analiz sonuçlarına göre metaverse ortamında bir eğitim platformu ve sınıf ortamı tasarlanmıştır.
- 1 July 2023
- Published by Elsevier in Pattern Recognition
- Vol. 143
Face age synthesis: A review on datasets, methods, and open research areas
- 1 July 2023
- Published by Elsevier in Pattern Recognition
- Vol. 143
Face age synthesis: A review on datasets, methods, and open research areas
- 28 August 2020
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 3 (2) , 99-111
Detection of Cervix Cancer from Pap-smear Images
Pap-smear test is used to detect cervical cancer, which ranks fourth in the ranking of cancer diseases in women worldwide. In this study, it is aimed to design a computer based decision system that can detect cervical cancer at an early stage. Normal and abnormal cells are found in the cervix images obtained as a result of the pap-smear test and the abnormal cells are marked on the image. The features extracted from the images were examined with pathologists and a dataset was created. For each of the 917 images in the Herlev dataset, these features were extracted and stored in a dataset. Support Vector Machines (SVM), Naive Bayes, Random Forest (RF), Multilayer Perceptron (MLP), Logistic Regression (LR), K- Nearest Neighbor (KNN) methods were applied to the created dataset, and accuracy values between 83% and 92% were obtained.
- 28 August 2020
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 3 (2) , 99-111
Detection of Cervix Cancer from Pap-smear Images
Pap-smear test is used to detect cervical cancer, which ranks fourth in the ranking of cancer diseases in women worldwide. In this study, it is aimed to design a computer based decision system that can detect cervical cancer at an early stage. Normal and abnormal cells are found in the cervix images obtained as a result of the pap-smear test and the abnormal cells are marked on the image. The features extracted from the images were examined with pathologists and a dataset was created. For each of the 917 images in the Herlev dataset, these features were extracted and stored in a dataset. Support Vector Machines (SVM), Naive Bayes, Random Forest (RF), Multilayer Perceptron (MLP), Logistic Regression (LR), K- Nearest Neighbor (KNN) methods were applied to the created dataset, and accuracy values between 83% and 92% were obtained.