Profile URL: https://hdl.handle.net/20.500.12573/2641
Job Title:Doç. Dr.
Email Address:rifat.kurban@agu.edu.tr
Main Affiliation:02. 04. Bilgisayar Mühendisliği
Status: Current Staff
ORCID:
0000-0002-0277-2210
0000-0002-0277-2210Scopus ID:
24729361200
24729361200YÖK Akademik: 8A7909B989B27EE0
Google Scholar:
wh5YHHgAAAAJ
wh5YHHgAAAAJWeb of Science ID:
B-1175-2012
B-1175-2012Name Variants:
Kurban, R. Kurban, Rifat
20 results
Scholarly Output Search Results
Now showing 1 - 10 of 20
Article Fine-Tuning Large Language Models for Turkish Flutter Code Generation(Sakarya University, 2025) Uluırmak, Buğra Alperen; Kurban, RifatThe rapid advancement of large language models (LLMs) for code generation has largely centered on English programming queries. This paper focuses on a low-resource language scenario, specifically Turkish, in the context of Flutter mobile app development. Two representative LLMs (a 4B-parameter multilingual model and a 3B code-specialized model) on a new Turkish question-and-answer dataset for Flutter/Dart are fine-tuned in this study. Fine-tuning with parameter-efficient techniques yields dramatic improvements in code generation quality: Bilingual Evaluation Understudy (BLEU), Recall-Oriented Understudy for Gisting Evaluation (ROUGE-L), Metric for Evaluation of Translation with Explicit Ordering (METEOR), Bidirectional Encoder Representations from Transformers Score (BERTScore), and CodeBLEU scores show significant increases. The rate of correct solutions increased from ~30–70% (for base models) to 80–90% after fine-tuning. The performance trade-offs between models are analyzed, revealing that the multilingual model slightly outperforms the code-focused model in accuracy after fine-tuning. However, the code-focused model demonstrates faster inference speeds. These results demonstrate that even with very limited non-English training data, customizing LLMs can bridge the gap in code generation, enabling high-quality assistance for Turkish developers comparable to that for English. The dataset was released on GitHub to facilitate further research in multilingual code generation.Conference Object Citation - Scopus: 4Re-Exploring the Kayseri Culture Route by Using Deep Learning for Cultural Heritage Image Classification Cultural Heritage Image Classification by Using Deep Learning: Kayseri Culture Route(Association for Computing Machinery, 2024-05-25) Kevseroğlu, Ozlem; Kurban, RifatThe categorization of images captured during the documentation of architectural structures is a crucial aspect of preserving cultural heritage in digital form. Dealing with a large volume of images makes this categorization process laborious and time-consuming, often leading to errors. Introducing automatic techniques to aid in sorting would streamline this process, enhancing the efficiency of digital documentation. Proper classification of these images facilitates improved organization and more effective searches using specific terms, thereby aiding in the analysis and interpretation of the heritage asset. This study primarily focuses on applying deep learning techniques, specifically SqueezeNet convolutional neural networks (CNNs), for classifying images of architectural heritage. The effectiveness of training these networks from scratch versus fine-tuning pre-existing models is examined. In this study, we concentrate on identifying significant elements within images of buildings with architectural heritage significance of Kayseri Culture Route. Since no suitable datasets for network training were found, a new dataset was created. Transfer learning enables the use of pre-trained convolutional neural networks to specific image classification tasks. In the experiments, 99.8% of classification accuracy have been achieved by using SqueezeNet, suggesting that the implementation of the technique can substantially enhance the digital documentation of architectural heritage. © 2024 Elsevier B.V., All rights reserved.Conference Object Plant Leaf Disease Segmentation Using U-Net++: A Comparative Evaluation with Classical Baselines(Institute of Electrical and Electronics Engineers Inc., 2026) Donmez, Huseyin Cahid; Kurban, Rifat; Ayaz, Vedat; Samet Varel, BurakConference Object Citation - WoS: 3Citation - Scopus: 4Fine Tuning DeepSeek and Llama Large Language Models with LoRA(IEEE, 2025) Uluirmak, Bugra Alperen; Kurban, RifatIn this paper, Low-Rank Adaptation (LoRA) finetuning of two different large language models (DeepSeek R1 Distill 8B and Llama3.1 8B) was performed using the Turkish dataset. Training was performed on Google Colab using A100 40 GB GPU, while the testing phase was carried out on Runpod using L4 24 GB GPU. The 64.6 thousand row dataset was transformed into question-answer pairs from the fields of agriculture, education, law and sustainability. In the testing phase, 40 test questions were asked for each model via Ollama web UI and the results were supported with graphs and detailed tables. It was observed that the performance of the existing language models improved with the fine-tuning method.Article Citation - WoS: 7Citation - Scopus: 8An Optimal Concentric Circular Antenna Array Design Using Atomic Orbital Search for Communication Systems(Walter de Gruyter Gmbh, 2024-05-06) Durmus, Ali; Yildirim, Zafer; Kurban, Rifat; Karakose, ErcanIn this study, optimum radiation patterns of Concentric Circular Antenna Arrays (CCAAs) are obtained by using the Atomic Orbital Search (AOS) algorithm for communication spectrum. Communication systems stands as a nascent technological innovation poised to revolutionize the landscape of wireless communication systems. It distinguishes itself through its hallmark features, notably an exceptionally high data transmission rate, expanded network capacity, minimal latency, and a commendable quality of service. The most important issue in wireless communication is a precision antenna array design. The success of this design depends on suppressing the maximum sidelobe levels (MSLs) values of the antenna in the far-field radiation region as much as possible. The AOS, which is a rapid and flexible search algorithm, is a novel physics-based algorithm. The amplitudes and inter-element spacing of CCAAs are optimally determined by utilizing AOS to the reduction of the MSLs. In this study, CCAAs with three and four rings are considered. The number of elements of these CCAAs has been determined as 4-6-8, 8-10-12 and 6-12-18-24. The radiation patterns obtained with AOS are compared with the results available in the literature and it is seen that the results of the AOS method are better.Article Citation - WoS: 2Citation - Scopus: 2Investigation of the Structural and Magnetic Properties of Rapidly Solidified Nd-Fe Alloys(Springer, 2024-07) Aytekin, Orkun; Kurban, Rifat; Durmus, Ali; Colak, Hakan; Karakose, ErcanThis study introduces the first literature report of rapidly solidified Nd-Fe-B-Ce alloys fabricated using the melt-spinning technique at varying disc rotation speeds. The resulting alloy images are then analyzed using various image processing techniques, and their structural and magnetic characteristics are described. The alloys are characterized using a variety of methods, including x-ray diffraction (XRD), field-emission scanning electron microscopy (FE-SEM), energy-dispersive x-ray spectroscopy (EDX), differential thermal analysis (DTA), vibrating sample magnetometry (VSM), and Vickers microhardness tests. By using XRD, the tetragonal hard magnetic Nd2Fe14B phase is detected in the Nd30Fe65B0.9Ce5 alloy. The FE-SEM microstructure analysis shows that the grain structure of the ingot alloy is indistinct, and the tetragonal symmetric structure begins to appear at disc rotation speeds of 20 m/s and 40 m/s. The analysis of FE-SEM images using histogram analysis, the image segmentation technique, and VSM method reveals that the coercivity values of the sample produced at the 80 m/s solidification speed increased by approximately 34% when compared to the ingot alloy.Conference Object Latency-Aware Speaker Identification in Browser Environments: A Comparative Study of Lightweight vs. Deep Architectures(Institute of Electrical and Electronics Engineers Inc., 2026) Cahid Donmez, Huseyin; Varel, Burak Samet; Kurban, Rifat; Ayaz, VedatArticle Multilevel Thresholding for Brain MR Image Segmentation Using Swarm-Based Optimization Algorithms(2024) Sahin, Omur; Kurban, Rifat; Toprak, Ahmet NusretBir görüntüyü bölüt adı verilen çeşitli piksel kümelerine ayırma işlemi olan görüntü bölütleme, görüntü işlemede önemli bir tekniktir. Görüntü bölütleme, görüntünün karmaşıklığını azaltmakta ve görüntüyü bölütlere ayırarak analiz edilmesini kolaylaştırmaktadır. Görüntü bölütlemenin en basit ancak etkin yollarından biri, piksellerin değerlerine göre birden çok bölgeye ayrıldığı çok düzeyli eşiklemedir. Bu çalışma, yaygın kullanılan sürü tabanlı optimizasyon algoritmalarının beyin MR görüntülerinde çok düzeyli eşikleme tabanlı görüntü bölütleme performansını araştırmayı ve karşılaştırmayı amaçlamaktadır. Yedi sürü zekâsı temelli optimizasyon algoritması: Parçacık Sürü Optimizasyonu (PSO), Yapay Arı Kolonisi (ABC), Gri Kurt Optimize Edici (GWO), Güve Alevi Optimizasyonu (MFO), Karınca Aslanı Optimizasyonu (ALO), Balina Optimizasyonu (WOA) ve Denizanası Arama Optimizasyon (JS) eşik seviyelerini belirlemek üzere beyin MR görüntülerine uygulanarak karşılaştırılmaktadır. Bahsi geçen algoritmalar ile yapılan deneylerde minimum çapraz entropi ve sınıflar arası varyans amaç fonksiyonları kullanılmıştır. Kapsamlı deneyler, JS, ABC ve PSO algoritmalarının daha iyi performans sergilediğini göstermektedir.Article Citation - WoS: 19Citation - Scopus: 21Investigation of the Performance and Properties of ZnO/GO Double-Layer Supercapacitor(Pergamon-Elsevier Science Ltd, 2024-08) Buyukkurkcu, Handan; Durmus, Ali; Colak, Hakan; Kurban, Rifat; Sahmetlioglu, Ertugrul; Karakose, ErcanComposite electrode material was formed by mixing reduced graphene oxide (rGO) and zinc oxide (ZnO) compound, using the Hummers and green synthesis methods, respectively. Of rGO powder, 10 g was mixed with 10%, 20% and 30% ZnO, and composite electrodes were obtained by using 10% binder. The energy storage performance and structural characteristics of the supercapacitor were evaluated by analyzing the capacitance values of the synthesized electrodes. The structural characterization of ZnO/rGO composites was performed using X-ray diffraction and field-emission scanning electron microscopy. The electrochemical properties of the ZnO/GO electrodes were analyzed by cyclic voltammetry, electrochemical impedance and galvanostatic charge -discharge tests. The specific capacitance value of electrodes increased as zinc content increased in the ZnO/ rGO composite material used to produce electrodes. The maximum specific capacitance values were measured at 5 mV/s scanning rate as 194.23 (rGO), 366.81 (10% ZnO), 383.18 (20% ZnO) and 410.48 F/g (30% ZnO). In conclusion, the use of composite material formed by the combination of ZnO nanoparticles obtained by green synthesis method from orange peel and graphene oxide increased the electrochemical efficiency of the supercapacitor.Conference Object Benchmarking CNN Architectures for Eye Disease Detection With Transfer Learning Techniques(Institute of Electrical and Electronics Engineers Inc., 2025-06-27) Keles, Tolgahan; Aykanat, Muhammet Ali; Kurban, RifatIn this study, convolutional neural networks (CNN)-based approaches were compared to classify eye diseases using transfer learning techniques. A series of data augmentation strategies, including random rotation, shifting, shearing, zooming, and horizontal flipping, were applied to increase the training data's robustness and diversity. Several state-of-the-art CNNs, including ResNet50, VGG19, EfficientNetB0, Xception, InceptionV3, DenseNet121, MobileNetV2, NASNetMobile, and ConvNeXtBase, were fine-tuned through transfer learning. During training, models were evaluated based on their accuracy, training time, and validation performance, while early stopping mechanisms were employed to prevent overfitting. Experimental results demonstrated that DenseNet121 achieved the highest validation accuracy (72%) during the training phase and the best test set performance with an accuracy of 68% and an AUC-ROC of 0.93. MobileNetV2, on the other hand, provided a strong balance between classification accuracy (65%) and low inference time (7.28 ms), making it appropriate for real-time uses. The findings highlight the importance of selecting appropriate architectures by considering both predictive performance and computational efficiency, particularly in the context of medical imaging, where real-world deployment constraints are critical. © 2025 Elsevier B.V., All rights reserved.