11 results
Search Results
Now showing 1 - 10 of 11
Conference Object Citation - WoS: 1An Undergraduate Curriculum for Deep Learning(Ieee, 2018) Tirkes, Guzin; Ekin, Cansu Cigdem; Sengul, Gokhan; Bostan, Atila; Karakaya, MuratDeep Learning (DL) is an interesting and rapidly developing field of research which has been currently utilized as a part of industry and in many disciplines to address a wide range of problems, from image classification, computer vision, video games, bioinformatics, and handwriting recognition to machine translation. The starting point of this study is the recognition of a big gap between the sector need of specialists in DL technology and the lack of sufficient education provided by the universities. Higher education institutions are the best environment to provide this expertise to the students. However, currently most universities do not provide specifically designed DL courses to their students. Thus, the main objective of this study is to design a novel curriculum including two courses to facilitate teaching and learning of DL topic. The proposed curriculum will enable students to solve real-world problems by applying DL approaches and gain necessary background to adapt their knowledge to more advanced, industry-specific fields.Conference Object Multi-Label Movie Genre Detection From Movie Posters Using Deep Learning Algorithms(Institute of Electrical and Electronics Engineers Inc., 2024) Yılmaz, A.A.In the field of cinema, the concept of genre has emerged as a concept that basically includes films that have common characteristics in terms of subject matter, have adopted a common method, and have a low risk of error because they have been tried before. Identifying the genres of movies is a challenging task because genres are intangible features that are not physically present in any movie scene, so off-the-shelf image detection models may not be easily integrated into this process. In this study, we aim to address the detection of movies according to their genres using deep learning algorithms. Movie poster data of IMDB and MM-IMDB datasets were utilized in our multi-label movie genre detection studies. In our experiments, we utilized four modern pre-trained models follow as DenseNet, VGG-16, ResNet-50, and MobileNet, and evaluated their performance using performance metric values such as accuracy, precision, recall, and F-score. According to the obtained empirical results, the DenseNet architecture achieved the highest accuracy values compared to other deep learning methods in detecting multi-label movie genre detection with an impressive rates of 91.64% and 92.56%. © 2024 IEEE.Conference Object Citation - WoS: 1Parking Space Occupancy Detection Using Deep Learning Methods(Ieee, 2018) Akinci, Fatih Can; Karakaya, MuratThis paper presents an approach for gathering information about the availabilty of the parking lots using Convoltional Neural Network (CNN) for image processing running on an embedded system. By using an efiicent neural network model, we made it possible to use a very low cost embedded system compared to the ones used in previous works on this topic. This efficient model's performance is compared to one of the models that proved its accuracy in image classification competitions. In these tests, we used datasets that has thousands of different images taken from parking lots in different light and weather conditions.Conference Object Citation - Scopus: 2Enhancing Image Resolution With Generative Adversarial Networks(Institute of Electrical and Electronics Engineers Inc., 2022) Yildiz,B.Super-resolution is the process of generating high-resolution images from low-resolution images. There are a variety of practical applications used in real-world problems such as high-definition content creation, surveillance imaging, gaming, and medical imaging. Super-resolution has been the subject of many researches over the past few decades, as improving image resolution offers many advantages. Going beyond the previously presented methods, Generative Adversarial Networks offers a very promising solution. In this work, we will use the Generative Adversarial Networks-based approach to obtain 4x resolution images that are perceptually better than previous solutions. Our extensive experiments, including perceptual comparison, Peak Signal-to-Noise Ratio, and classification success metrics, show that our approach is quite promising for image super-resolution. © 2022 IEEE.Conference Object Citation - Scopus: 8Parking space occupancy detection using deep learning methods;(Institute of Electrical and Electronics Engineers Inc., 2018) Akinci,F.C.; Karakaya,M.This paper presents an approach for gathering information about the availabilty of the parking lots using Convoltional Neural Network (CNN) for image processing running on an embedded system. By using an eflicent neural network model, we made it possible to use a very low cost embedded system compared to the ones used in previous works on this topic. This efficient model's performance is compared to one of the models that proved its accuracy in image classification competitions. In these tests, we used datasets that has thousands of different images taken from parking lots in different light and weather conditions. © 2018 IEEE.Article Revolutionizing Glaucoma Care: Harnessing Artificial Intelligence for Precise Diagnosis and Management(Gazi Eye Foundation, 2025) Ucgul, A.Y.; Aktaş, Z.Glaucoma is a leading cause of irreversible blindness worldwide, necessitating early detection and effective management to prevent vision loss. Recent advancements in artificial intelligence (AI) have revolutionized glaucoma care by enhancing diagnostic accuracy, monitoring disease progression, and personalizing treatment strategies. AI models, including machine learning and deep learning algorithms, have demonstrated exceptional performance in analyzing fundus photography, optical coherence tomography, and visual field data, surpassing traditional diagnostic methods. Convolutional neural networks have shown high sensitivity and specificity in detecting glaucomatous changes, while vision transformers and hybrid AI models further refine risk assessment and prognosis. Additionally, AI-powered monitoring systems utilizing multi-modal data integration allow for more precise prediction of disease progression and the need for surgical intervention. The incorporation of AI into telemedicine and wearable intraocular pressure sensors extends glaucoma management to remote and underserved populations. Despite these advancements, challenges remain, including issues related to algorithm generalizability, data standardization, bias, and ethical concerns regarding AI-driven clinical decision-making. To maximize AI’s potential in glaucoma care, further interdisciplinary research, regulatory oversight, and multi-center validation studies are needed. By addressing these challenges, AI can be effectively integrated into clinical practice, leading to improved early detection, enhanced treatment strategies, and more personalized patient care. The future of AI in glaucoma management holds great promise, paving the way for a more data-driven and patient-centered approach to combating this sight-threatening disease. © 2024 The author(s).Conference Object Citation - Scopus: 2An Undergraduate Curriculum for Deep Learning(Institute of Electrical and Electronics Engineers Inc., 2018) Tirkes,G.; Ekin,C.C.; Engul,G.; Bostan,A.; Karakaya,M.Deep Learning (DL) is an interesting and rapidly developing field of research which has been currently utilized as a part of industry and in many disciplines to address a wide range of problems, from image classification, computer vision, video games, bioinformatics, and handwriting recognition to machine translation. The starting point of this study is the recognition of a big gap between the sector need of specialists in DL technology and the lack of sufficient education provided by the universities. Higher education institutions are the best environment to provide this expertise to the students. However, currently most universities do not provide specifically designed DL courses to their students. Thus, the main objective of this study is to design a novel curriculum including two courses to facilitate teaching and learning of DL topic. The proposed curriculum will enable students to solve real-world problems by applying DL approaches and gain necessary background to adapt their knowledge to more advanced, industry-specific fields. © 2018 IEEE.Article Citation - WoS: 2Citation - Scopus: 4Precision Forecasting for Hybrid Energy Systems Using Five Deep Learning Algorithms for Meteorological Parameter Prediction(Elsevier Sci Ltd, 2025) Ceylan, Ceren; Yumurtaci, ZehraThe intermittent nature of renewable energy sources necessitates accurate power production forecasting to ensure system sustainability and balance between energy supply and demand. Although the deep learning-based meteorological forecasting is significantly studied in literature, most of the current literature applies single-algorithm based on each individual energy source and less multi-algorithm based on comparative studies on multiple architectures as applied to integrated hybrid systems. In addition, most of the research uses the same algorithmic solution to all the meteorological parameters without identifying parameter-specific optimization potential, and recent research is verified on actual future time steps instead of historical train-test split. This study presents a comprehensive comparative analysis of five deep learning algorithms, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and CNN-LSTM hybrid, for forecasting critical meteorological parameters (wind speed, ambient temperature, and solar radiation) that determine energy output in a wind and solar-based hybrid energy system (HES). Using five years of Istanbul meteorological data (2018-2022), optimal algorithms were systematically identified for each parameter through rigorous hyperparameter optimization and cross-validation. Key results demonstrate that GRU achieves superior performance in wind speed prediction (RMSE: 0.049 m/s, R2: 0.8634) and solar radiation forecasting (RMSE: 0.146 W/m2, R2: 0.6643), while CNN-LSTM excels in ambient temperature prediction (RMSE: 0.011 degrees C, R2: 0.9976). The integrated approach predicted annual hybrid system energy production with 89 % accuracy, demonstrating 0.48 % deviation from observed values. Most significantly, our framework successfully forecasted sixth year (2023) energy production with 1.55 % error, validating its real-world applicability. This research contributes to the methodological advancement of renewable energy forecasting by systematically identifying optimal algorithmic approaches for different meteorological parameters in hybrid systems, thereby supporting the integration of intermittent renewable sources into sustainable energy infrastructures.Article Univariate Deep Learning Models for Short-Term Electricity Load Forecasting from Renewables(Ankara Univ, Fac Sci, 2025) Kabran, Fatma Basoglu; Unlu, Kamil DemirberkRenewable energy offers a cost-effective, carbon-free solution for energy needs, while protecting the environment. Accurate forecasting of electricity generation from renewable sources is crucial for the efficiency of modern power grids. This study employs a univariate deep learning approach to predict daily renewable energy generation, evaluating Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) as candidate models. Five performance metrics-mean absolute error, root mean squared error, mean absolute percentage error, mean absolute scaled error and the coefficient of determination-are employed to assess the forecasting power of the algorithms. The empirical results show that CNN outperforms other models, achieving an R2 of almost 94%. This research shows that the univariate model based on historical data of electricity load generated from renewables can accurately predict day-ahead electricity load, even without meteorological data.Conference Object An Undergraduate Curriculum for Deep Learning(Institute of Electrical and Electronics Engineers Inc., 2018) Tirkes,G.; Ekin,C.C.; Engul,G.; Bostan,A.; Karakaya,M.Deep Learning (DL) is an interesting and rapidly developing field of research which has been currently utilized as a part of industry and in many disciplines to address a wide range of problems, from image classification, computer vision, video games, bioinformatics, and handwriting recognition to machine translation. The starting point of this study is the recognition of a big gap between the sector need of specialists in DL technology and the lack of sufficient education provided by the universities. Higher education institutions are the best environment to provide this expertise to the students. However, currently most universities do not provide specifically designed DL courses to their students. Thus, the main objective of this study is to design a novel curriculum including two courses to facilitate teaching and learning of DL topic. The proposed curriculum will enable students to solve real-world problems by applying DL approaches and gain necessary background to adapt their knowledge to more advanced, industry-specific fields. © 2018 IEEE.
