7
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Discipline
Özgül öğrenme güçlüğü olan çocukların eğitsel alıştırmaları için uyarlanabilir ciddi oyunlar geliştirilmesi
Recent developments in technology and the video game industry enabled video games to play essential roles, especially in education, health, and military services under the umbrella term serious games (SGs). Recent studies show that playing computer games is common among children and teenagers, and SGs ease the learning process by providing immersive and interactive platforms. Recently, SGs have been used as an alternative method in education and diagnosis of children with a specific learning difficulty (SpLD). However, these studies have focused on only one or two types of SpLD at the same time. To analyze a broader range of SpLD, in this thesis, five different SGs were designed and developed in order to be used in the training of children with any SpLD. Usability tests were applied to two different participant groups —10 educators and 25 students with SpLD— while they played the games during which think-aloud protocols were implemented. In the second part of this study, a rule-based adaptive difficulty level enhanced the two of the five games, and these modified games were played and tested by the students. The results show that both participant groups had a positive attitude on using SGs together with a complete training tool for students with any type of SpLD. Also, students performed better when they played the adaptive difficulty levels since the games were adjusted automatically due to their performance.
Manüel öznitelik çıkarımı ve derin öğrenme kullanılarak kumaş yumuşaklığı ve boncuklanma değerlerinin objektif bir şekilde ölçülmesi ve sınıflandırılması
Fabric softness is a complex tactile sensation perceived by the user even before the fabrics are worn. Softness is usually the property of surface perceived by touching or pressing a finger on the fabric surface. Fabric friction properties significantly affect the tactile sensation of the garments. The yarn used, the finishing works, and the fabric structure (weaving, knitting, etc.) affect the softness. In addition, the hardness of the water used during washing, washing movements, the amount and content of the detergent and softener used also have permanent effects on the fabric softness. Softness can be evaluated by the jury members with proven effectiveness according to the predetermined scale. Our achievement within the scope of the thesis is to eliminate the differences that may occur as a result of the subjective evaluation, which may arise from qualitative observations by basing the degree of softness evaluated qualitatively on numerical data and to obtain clearer and more precise results by adding quantitative features to the evaluation process. The methodology developed for softness assessment is also applied for another textile deterioration parameter, namely pilling, and its results are also reported.
Küçük ölçekli verilerde araç tespiti için üretken metodlarla veri artırma
Scarcity of training data is one of the prominent problems for deep neural networks, which commonly require high amounts of data to display their potential. Data augmentation techniques are frequently applied during the pre-training and training phases of deep neural networks to overcome the problem of having insufficient data for training. These techniques aim to increase a neural network's generalization performance on unseen data by increasing the number of training samples and provide a more representative distribution to the system during training. In this work, we focus on improving vehicle detection in aerial images by proposing a data augmentation method that does not need any extra supervision than the bounding box annotations of the vehicle instances in the training data. The methods we used are based on a conditional Generative Adversarial Network (cGAN). The proposed method is not exclusive and can be used in association with classical augmentation techniques to further improve object detection performance. We showed that the proposed data augmentation method increases the Average Precision by up to 25.2%, 32.7%, and 25.7% when integrated with Pluralistic, PSGAN, and DeepFill respectively.
Gürültü altuzayı tabanlı DOA kestirim algoritmalarının CPU ve GPU üzerinde parallelleştirilmesi
Direction-of-Arrival (DOA) estimation is known as an active research area, and it is studied under array signal processing. The algorithms in this area are widely used in various applications such as sonar, search-and-rescue, navigation, and geolocation. However, achieving a real-time system performance is sometimes a challenging task for these algorithms. In this thesis, four noise subspace-based DOA estimation algorithms (PHD, MUSIC, EV, and MN) were considered and implemented in MATLAB, C/C++, and CUDA. MATLAB implementations were mainly used for theoretical and numerical analyses. Whereas, C/C++ implementations were initially used for constructing the parallelization structure and they were realized in two versions working serially and in parallel manner (via OpenMP). Within the scope of theoretical analysis, these algorithms were compared with each other in terms of DOA estimation accuracy and effects of change in different parameters (e.g., array geometry, array aperture, SNR level, etc.) on the accuracy were observed. On the other hand, in terms of implementation-based experiments, all the codes in MATLAB, C/C++, and CUDA were evaluated from both numerical and performance viewpoints. The general numerical validation of C/C++ and CUDA codes was realized against ground-truth MATLAB codes. After the initial assessment of the CUDA code performance, some GPU-based optimizations were applied and the corresponding performance improvements were evaluated. The CUDA code performance was benchmarked on the PC platforms with different system configurations. Consequently, a considerable speedup was achieved for CUDA codes compared to baseline multi-threaded C/C++ codes.
Yeni bir ilişkisel akıl yürütme test ortamından belirgin ilişkisel bilgileri bir öğrenme ajanı ile çıkarmak
In recent studies, reinforcement learning (RL) agents work in ways that are specialized according to the tasks, and most of the time, their decision-making logic is not interpretable. By using symbolic artificial intelligence techniques like logic programming, statistical methods-based agent algorithms can be enhanced in terms of generalizability and interpretability. In this study, the PrediNet architecture is used for the first time in an RL problem, and in order to perform benchmarking, the multi-head dot-product attention network (MHDPA) was used. By using the PrediNet module, relational information among the objects in the environment can be extracted explicitly. This information is in a form that can be processed in logic programming tools, and the network becomes more interpretable. In order to measure the relational information extraction performances of these two methods, a new test environment, relational-grid-world (RGW), is developed. RGW environment can be generated procedurally from objects with different features, pushing the agent to make complex combinatorial selections in this environment. In the performed tests and the RGW environment, a baseline environment called Box-World is used for comparing both environments and networks separately. The results show that both MHDPA and PrediNet architecture have similar performances in both environments, and the RGW environment is able to measure the relational reasoning capacity of the networks.
Cbrn-e eğitimleri için ciddi oyunlar geliştirilmesi: Bilgisayar ve sanal gerçeklik platformlarında karşılaştırmalı çalışma
The resolution of a crisis and dealing with the consequences that come afterward of these events requires comprehensive training of authorized individuals. These training sessions consist of executing a series of different tasks based on a pre-defined crisis scenario. The main challenge of performing such an exercise is reusability. Repeating sessions can be costly in terms of money and time. Terrorism acts such as Chemical, Biological, Radioactive, Nuclear, and explosive (CBRN-e) attacks are classified among main crises and become a focused task across the globe. In this thesis, two CBRN-e based scenarios that are introduced in the EU H2020 European Network of CBRN Training Centers (eNotice) project's joint activities —France and Belgium— are implemented as serious games with the motivation of creating cost-effective and repetitive digital training sessions. The scenarios have different elements such as crisis management in hospitals, deployment of related units to the crime scene, objectives of individuals such as doctors, nurses, and investigation teams. Evaluation of the study has been executed by 16 CBRN-e experts from the eNotice that are also participated in the joint activities in France and Belgium. Experts played the games —both in Desktop and Virtual Reality— and answered questionnaires about presence, system usability, immersive tendency, technology acceptance model. Answers to the questionnaires and feedback indicated that enjoyment of the games had enabled users to enhance their learning while having a good time. Furthermore, Virtual Reality has much more improved the feeling of "presence" and created a unique experience, according to the participants.
Protein etkileşimlerinin sanal gerçeklik ve karma gerçeklik ortamlarında görselleştirmesi
Protein-protein interactions (PPI) define the physical contact of two or more protein structures. When these interactions are combined, the protein-protein interaction network (PPIN) is formed. The interactions between protein structures are distinct interactions—they happen in specific binding locations on proteins, and they have a specific biological function that they take on. With these networks, the processes within a cell or a living organism when healthy or diseased can be studied. In this thesis, a 3D visualization framework to envisage PPIs and PPINs in virtual reality (VR) and mixed reality (MR) environments will be developed. Detailed performance tests of the framework will be provided and analyzed for both VR and MR environments.