113
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Discipline
An improved transfer learning based siamese network for face recognation
In the digital era, interest in algorithms and theories for face recognition systems FR have been growing rapidly. Criminal identification, video surveillance, unmanned and autonomous vehicles, and building access control, are just examples of real applications that are gaining attraction among industries. FR is currently a highly difficult and complex subject in deep learning DL, neural networks, pattern analysis, and computer vision domains. Different learning groups, including controller environment and uncontrolled environment, have debated this issue. FR is a novel artificial intelligence application that DL has discovered recently. Earlier, many efforts have been dedicated to building accurate and adaptive FR models. However, recognizing faces in unconstrained environments poses a significant challenge due to various factors such as head pose, age, illumination, and facial expression variations, and others. Therefore , the aim of this study is to develop an efficient FR approach based on a Siamese neural network SNNs and Transfer Learning methods TL. The proposed approach employs SNNs with Visual Geometry Group 16 VGG-16 as a background for efficient FR especially in the case of individuals having similar facial features, and for the purpose of precisely identifying individuals in varying environments. The proposed approach consists of several phases, first data gathering, second data pre-processing, third model building, then comparing the proposed VGG-16 with three more convolutional networks (EfficientNet, ResNetB0, and ConvNeXt algorithms) to keep sure that the proposed approach is robust. For this purpose, labeled faces in the wild LFW dataset were used for SNN with VGG-16. After training the networks, the SNNs with VGG-16 exhibited low loss and a high accuracy in FR. Performance of the architectures were measured using K-Fold cross validation for 5 partition. According to results, EfficientNet, RestNet50 and ConvNext produced 77.75% accuracy, 95% and 93.75 % accuracy respectively.On the other hand, SNN with VGG-16 exhibited a low loss and produced the best accuracy in FR with 96.25%.
Yükseköğretim çalışmalarında entegre edilmiş harmanlanmış öğrenme teknolojisinin etkisi
Anahtar kelimeler: e-Sınıf, e-Öğrenim, e-İçerik, Yükseköğrenimde e-öğrenme, M-öğrenme, sosyal öğrenme, u-Öğrenme ve harmanlanmış Öğrenme. Günümüzde, eğitim teknoloji ve internet kullanımıyla önemli değişimler yaşamakta, yeni öğretim ve öğrenim yöntemleri geliştirilmektedir. Bilgi edinimini teşvik etmek için yaygın şekilde kullanılan öğretim yöntemlerinden biri de çeşitli formatlarda mevcut olan sanal ortamların kullanımını kapsamaktadır. Çevrimiçi olarak erişim sağlanabilecek öğretim-öğrenim sistemleri bu ortamlara örnek olarak gösterilebilir. Bu araştırma özellikle, çevrimiçi öğrenme içeriğinin tasarımı ve uygulanması için BİT araçlarını kullanarak öğrenim ve öğretim materyalleri sunmaktadır. Bu durum aşağıda belirtilenlere odaklanmış üç temel alanı kapsamaktadır; 1.) e-Kitap kullanarak etkileşimli multimedya ders içeriği (e-İçerik) tasarımı ve uygulaması; 2.) Karma eşzamansız öğrenmeye entegre edilmiş ÖYS, e-Ödev, e-Sınav tasarımı ve uygulaması; 3.) Etkileşimli e-Sınıf oluşturumu ve uygulaması. Hem eşzamanlı hem de eşzamansız teknolojilere dayalı, entegre edilmiş karma çevrimiçi ders, eğitimcilerin yüz yüze öğrenmeyi desteklemelerinin yanı sıra, çevrimiçi öğrencilere işbirlikli faaliyetler ve öğrenme kaynakları sunmalarına da imkân tanımaktadır. Sakarya Üniversitesi Fen Edebiyat Fakültesi Fizik Bölümü'nde eğitim gören örnek grubundaki 122 öğrenci rastgele seçilmiştir. Veri toplamak için 40 soru içeren bir anket oluşturulmuş ve anketten alınan sonuç SPSS programı kullanılarak analiz edilmiştir. Çalışma kapsamında, e-Kitap kullanımına ilişkin elde edilen ilk bulgular şu şekildedir; Öğrencilerin çoğunluğu (%57.4'si) internet erişimi için akıllı telefonlarını kullanmaktadır, e-Kitap üzerinde geçirilen zamana yönelik incelemeler öğrencilerin %52.2'inin bu hususta bir saatlik bir zaman harcadığını ortaya koymuştur. Yapılan veri analizi "Katılıyorum" ve "Kesinlikle katılıyorum" şıklarının ilgili ankette en çok işaretlenen şıklar olduğunu göstermiştir. ÖYS platformunun kullanımına ilişkin elde edilen bulgular, öğrencilerin %59'unun internet erişimi için akıllı telefonlarını kullandığını ve %57.4'inin ÖYS platformu üzerinde bir saatlik bir zaman geçirdiğini göstermiştir. Yapılan veri analizi "Katılıyorum" ve "Kesinlikle katılıyorum" şıklarının ilgili ankette en çok işaretlenen şıklar olduğunu ortaya çıkarmıştır. e-Sınıf sisteminin 18 öğrenci kullanımına ilişkin elde edilen bulgular ise öğrencilerin %66.7'sinin internet erişimi için akıllı telefonlarını kullandığını göstermiştir. Yapılan veri analizi "Katılıyorum" ve "Kararsızım" şıklarının ilgili ankette en çok işaretlenen şıklar olduğunu ortaya koymuştur. Çalışmadan alınan tüm bu sonuçlar, daha fazla öğrencinin bulut ve HTML5 teknolojileri kullanarak sesli ve görüntülü derse daha kolay erişim ve paylaşım imkânı sunan çevrimiçi
Ölçülen klmyasal gazlarln makina ögrenmesi ile sınıflandırması
Bu tez çalışmasında amaçı endüstri 4.0'ın temel konularından olan gömülü sistem yazılım ve donanımlarını kullanarak elde edilen sensör verilerini makine öğrenmesi algoritmaları ile sınıflandırmak ve sürdürülebilir üretim için gerekli otomasyonun sağlanması için alt yapı oluşturmaktır. Günümüzde endüstride kullanılan tüm aygıtlar akıllı hale geliyor ve üretim sürecinin tüm seviyelerinde üretilen dijital veriler ürün kalitesini, esnekliğini ve fonksiyonelliğini artırmak için kullanılmaktadır. Üretim tesislerinden gelen veri sensörlerle toplanmaktadır. Büyük veri tifadesi ile bahsedilen bu verilerin insan kullanıcıların değerlendire bilmesi mümkün değildir. Bu veriyi makine öğrenmesi algoritmaları ile değerlendirmek, anlamlandırmak ve sürdürülebilir üretim için kullanmak ancak mümkün olabilir. Bu çalışmada endüstride gerek insan hayatını doğrudan etkilemesi ile gerek ise birçok ürünün üretiminde ortaya çıkması veya kullanmasından dolayı gazlar ın algılanması ve sınıflandırılması üzerinde durulmuştur. Gaz sensörleri, insan sağlığını ve özelliklerini tehdit eden yanıcı, yakıcı ve toksik gazları tespit etmek için endüstride ve yangınla mücadelede yaygın olarak kullanılmaktadır. Önce ardunia ile b gaz verilerini toplama ve makine öğrenmesi ile sınıflandırılması üzerine çalışılmıştır. Gazlar ile ilgili yeterince veri elde edilemediği için altı farklı uçucu organik gaz için bir veri kümesi veri seti indirilerek onun üzerinde çalışmalar yapılmıştır. Bu veri setinde; Amonyak, Asetaldehit, Aseton, Etilen, Etanol ve Toluen gazları bulunmaktadır.Üç yıllık bir süre boyunca, 16 metal oksit gaz sensörü kullanılarak elde edilmiştir. Bu veriler 13910 ölçüm ve 129 özellik içerir.Çalışmamızda bu verileri sınıflandırmak için çok sayıda makine öğrenmesi algoritmaları kullanılmıştır. Bunlar; k-en yakın komşular, Destek vektör makinesi, Rastgele orman ve Lojistik regresyon dur. Verileri sınıflandırmak için bu makine öğrenme algoritmaları Weka programında gerçekleştirilmiştir. Yapılan çalışmalar sonucunda k-en yakın komşular algoritmasının en iyi sonucu verdiğini ve sınıflandırma başarısı% 99,48dır.
Ethercat tabanlı içme suyu sistemi üzerinde MITRE ICS saldırı simülasyonu ve tespiti
Endüstriyel kontrol sistemleri, içerdiği teknoloji ve protokol çeşitliliğinden dolayı karmaşık sistemlerdir. Ancak kritik sistemler olduğundan olası herhangi bir saldırı durumunda yıkım etkisi de aynı oranda büyük olmaktadır. Bu yüzden kritik alt yapıların siber saldırılara karşı korunması ve sürekli izlenebilirliği önemli ve gereklidir. EKS çalışma yapısı olan OT, standart bilişim alt yapısından farklı performans ve güvenlik gereksinimlerine sahiptir. EKS sistemler, operasyonel süreçlerin gerçekleştiği saha cihazları ve bu cihazların yönetimini sağlayan kontrol sistemlerinden oluşmaktadır. Saldırganlar kontrol katmanından erişim sağladıktan sonra bütün sürece dahil olmaktadırlar. Bu durumun sonucu olarak kritik alt yapı sistemleri siber saldırılara karşı tehdit altındadır. Dolayısıyla sürekli izleme ve güvenlik denetimleri, kritik alt yapılar için de gerekli bir süreçtir. Bu çalışmada, kritik alt yapılardan biri olan su yönetim prosesi üzerinde siber saldırı ve tespit sistemine yönelik çalışmalar yapılmıştır. EtherCAT tabanlı çalışan su yönetim prosesi üzerinde saha cihazlarına yönelik toplamda 6 farklı saldırı vektörü MITRE ICS ATT&CK matrisindeki tekniklere göre geliştirilmiş ve bu saldırılar ağ trafiğinden elde edilen veriler ayrıştırılarak SVM algoritmasıyla tespit edilmiştir. Aynı proses üzerinde mühendislik bilgisayarı aracılığıyla kontrol merkezindeki SCADA sistemine yapılan saldırılarda ise MITRE ICS ATT&CK matrisinde bulunan 7 farklı teknik seçilerek saldırı senaryoları oluşturulmuştur. SCADA sistemine yönelik saldırı tespit sistemi için Wazuh HIDS kullanılmıştır. Her iki saldırının görselleştirilmesi ELK üzerinde yapılmıştır. Anahtar kelimeler: Endüstriyel kontrol sistemleri, saldırı tespit sistemi, MITRE ICS ATT&CK Matrisi
Bilgisayar gÖrmesine dayalı ve evrişimsel sinir ağları kullanarak yaş ve cinsiyet tahmı̇nı̇
İnsan yüzü; ifade, etnik kimlik, cinsiyet ve yaş gibi, kişi hakkında çok önemli niteliklere ve bilgilere sahiptir. İnsanoğlu bu bilgileri kolayca tespit edip analiz edebilir; örneğin insanların çoğu cinsiyet gibi insan özelliklerini kolayca tanıyabilir, kişinin sadece yüzünü görerek kadın ya da erkek olup olmadığını söyleyebilir. Aynı şekilde, kişinin yaşını belirleyebilir ve bu kişinin çocuk mu yoksa yetişkin mi olduğunu belirleyebilir. Öte yandan, insanları yüzlerinden tanımlamak, yaş ve cinsiyet bilgilerini çıkarmak için uygulamalar oluşturmak, modern dünyanın ihtiyaçları nedeniyle günlük hayatımızın birçok önemli alanında bağımlı olduğumuz bilgisayar vizyonu için zorlu bir görevdir. Bu çalışma, insanların yüzlerinden yaş ve cinsiyet tahminine yönelik derin bir öğrenme çözümüne odaklanmaktadır. Evrişimsel Sinir Ağları (CNN-Convolutunel Neurel Network) kullanarak IMDB-Wiki Veri Kümesi'nden transfer öğrenme ve ince ayar gerçekleştirilmektedir. İlk olarak, transfer modeli üç model olarak kullanılmıştır: Birinci model: Eğitilmesi daha hızlı olan ve görüntü sınıflandırması için kullanılan ImageNet veri kümesi ile önceden eğitilmiş olan MobileNet V2 modeli. İkinci model: Bir CNN türü olan Inception V3 modeli. Bu model Birçok evrişim ve maksimum havuzlama katmanından oluşur. Her iki model de Keras uygulamasidir. Ve son model bazı değişiklikleri kapsayan SSR- Net mimarisidir.
Üç boyutlu konumlandırma problemleri için yeni bir veri eşleştirme çözüm yöntemi: One-poınt ransac wıth epıpolar constraınt
The problem of Localization or Simultaneous Localization and Mapping has received a great deal of attention within the robotics literature, and the importance of the solutions to this problem has been well documented for successful operation of autonomous agents in a number of environments. Of the numerous solutions that have been developed for solving the problems, many of the most successful approaches continue to either rely on, or stem from noise ltering techniques, especially the Extended Kalman Filter method or Particle Filtering methods. Localization problems are downgraded to a data association problem after using mentioned lters. This topic has also received a great deal of attention in the robotics literature in recent years, and various solutions have been proposed. In the thesis, rst mostly studied methods, such as Joint Compatibility, Sequential Compatibility Nearest Neighbor, Joint Maximum Likelihood, one point RANSAC and epipolar consistency, are studied. As the second part of the thesis a new method is presented. One-Point RANSAC with Epipolar Constraint (OPRF) is based on RANSAC and epipolar geometry. Later the performance and consistency of the method will be compared to epipolar consistency solution.
Oyunlaştırılmış ders hazırlama sistemi
Good games are embedded with many elements that make them attractive to play. If lessons could also be gamified, they would cease to be boring and become attractive as well. This could result in students spending more time studying and learning in an efficient way. In this thesis, we present a method and tools for preparing gamified lessons. In addition to gamifying the delivery of lectures, assignments, exams, and grading, we incorporate many different game elements into lessons in order to improve the gamification of lessons. We present a visual authoring environment to guide instructors in preparing gamified lessons and also define an index to measure the "gameness" of a gamified lesson. An instructor could use this index to determine whether he/she has gamified his/her lesson enough. Furthermore, we present an interactive web application for students so that they can "play" the gamified lessons.
F5 sinirsel aktivite verisinden kol kinematiğinin gerçek-zamanlı çözme
Extending our knowledge about brain mechanisms and behavior can lead to many advantages and inspiration in the diagnosis of nervous system diseases and robotics and artificial intelligence. Ventral premotor cortex, i.e. area F5, in a macaque monkey's brain is one of the areas of interest in the literature. Studies have shown that F5 area in monkeys is involved in arm movements and hand configuration, enabling the animal to grasp objects with different shapes (different grip types). Furthermore, it is shown in the studies that F5 area contains neurons called mirror neurons which are active not only during the period the animal moves his arm and hand but also while the animal is observing another monkey or person performing the same action. In this study, we aim to investigate whether, by using F5 area neural activity, monkey's arm kinematics can be decoded in real-time or not. Furthermore, how the decoding capacity of mirror and non-mirror neurons can be differentiated. To this end, the neural behavior of 32 neurons (including both mirror and non-mirror neurons) in the stated area was recorded while a monkey was performing grasping tasks on different objects. Also, monkey's motion was video captured simultaneously. Using image processing techniques and tools, kinematics data was extracted from the videos. Later, the possibility of single neuron's decoding of the kinematics data was investigated. Results reveal that although single neuron real-time decoding of the kinematics is not always ideal, reasonable performance is achievable with selected neurons from both groups. Based on the results of this study non-mirror neurons seem to act as better single-neuron decoders. Although it seems obvious that population-level activity is required for more robust decoding, the single-neuron decoding accuracy can be considered as possible criteria to categorize neurons in the F5 area.
Model bazlı yazılım ürün hattı testi için araç desteği
We introduce a tool for automated adaptation of test models to be reused for a prod uct family. Test models are specified in the form of hierarchical Markov chains. They represent possible usage behavior regarding the features of systems as part of the product family. A feature model documents the variability among these features. Optional and alternative features in this model are mapped to a set of states in test models. These features are selected or deselected for each product to be tested. Transition probabilities on the test model are updated by our tool according to these (de)selections. As a result, the test case generation process focuses only on the se lected features. We conducted two controlled experiments, both in industrial settings, to evaluate the effectiveness of the tool. We used systems as part of digital TV and wireless access point(WAP) systems. For DTV systems 10 and for wireless access points 5 participants were involved in testing these systems, respectively. We mea sured the effort spent by each participant for the same set of tasks when our tool is used and when it is not. We observed that the tool reduces costs significantly. We also observed that the initial cost for adopting product line testing is amortized even for small product families with 13 DTV and 11 WAP products, respectively.
Gelişmiş taşıma ile HTTP üzerinden adaptif iletim
QUIC (Quick UDP Internet Connections) is an experimental and low-latency transport network protocol proposed by Google, which is still being improved and specified in the IETF. The viewer's quality of experience (QoE) in HTTP adaptive streaming (HAS) applications may be improved with the help of QUIC's low-latency, improved congestion control and multiplexing features. In this master thesis, we measured the streaming performance of QUIC on wireless and cellular networks in order to understand whether the problems that occur when running HTTP over TCP can be reduced by using HTTP over QUIC. The performance of QUIC was tested in the presence of network interface changes caused by the mobility of the viewer. We observed that QUIC resulted in quicker start of media streams, better streaming and seeking experience, especially during the higher levels of congestion in the network and had a better performance than TCP when the viewer was mobile and switched between the wireless networks. Furthermore, we investigated QUIC's performance in an emulated network that had a various amount of losses and delays to evaluate how QUIC's multiplexing feature would be beneficial for HAS applications. We compared the performance of HAS applications using multiplexing video streams with HTTP/1.1 over multiple TCP connections to HTTP/2 over one TCP connection and to QUIC over one UDP connection. To that effect, we observed that QUIC provided better performance than TCP on a network that had large delays. However, QUIC did not provide a significant improvement when the loss rate was large.
Otonom sütur için doku kesiği, sütur iğnesi, ve sütur ipliğinin çoklu kamera akışında görsel yerelleştirilmesi
Autonomous suturing systems could potentially replace the surgeon in suturing and benefit both patient and surgeon. Automating the suturing process involves many complex challenges and this paper proposes a comprehensive solution to the localization problems. In this context, localization problems refer to the challenges involved in the visual tracking of suture needle, suture thread, and tissue, which are fundamental processes for autonomous suturing that must be robust, precise, and computationally affordable. Surgical instruments' reflective, superfine, or non-rigid structures make their localization a complex problem. The contribution of this study is that it provides a combined algorithm for suture needle and thread detection, an algorithm for tissue cut detection. The overall localization approach proposed here defines the suture needle and its thread with fewer parameters, enabling a more precise and less computationally intensive localization. For the testing of the proposed approach, an experimental setup featuring two 7-DOF robots equipped with surgical instruments and two fully calibrated cameras were utilized. The localization experiments were done using realistic artificial (silicone) tissues and threaded surgical needles. The results of the experiments validated that the proposed method meets the requirements for performing automated robotic sutures.
Birlikte evrimleşme endeksi: Evrimsel bağlaşımı izlemek için bir metrik
This thesis proposes a new metric, namely the coevolution index (CEI), for measuring the relative evolutionary coupling of modules of a software system. CEI is inspired by the h-index, which is a popular metric used for measuring the productivity and citation impact of scholars and scientists. CEI of a module is equal to n, which is the number of times it is modified together with at least n other modules of the system. We develop a script that can calculate CEI for source files in a code repository. We analyze the repository of 7 software systems. Source files that are subject to a high number of changes to address issues tend to have high CEI scores. CEI also reflects a relative footprint in maintenance efforts by definition. Hence, it can help in tracking technical debt interest and focusing the refactoring efforts for improving maintainability and reusability.
Nesnelerin interneti tabanlı soğuk zincir lojistiği yönetim sisteminin modellenmesi ve benzetimi
The Dramatic growth of world economy results growth in the supply chain which demands logistics service to be agile, flexible and responsive in the face of uncertainty, especially for temperature sensitive products that need to be monitored and managed in the cold chain. To achieve this, Logistics companies must be supported by appropriate information technologies. Internet provides an effective means of driving information between customer and logistics provider, however, existing gap between products flow and information flow in logistic service has created a problem in getting real-time information about temperature sensitive items which make logistics management more challenging for decision makers. The growth of internet of things (IoT) gives a potential solution for monitoring, managing, and achieving real-time visibility and sharing information with the appropriate level of intelligence in cold chain industries. This paper demonstrates IoT enabled cold chain logistics that helps to enhance the decision support of all actors through managing, monitoring the real-time ambient temperature of the cold chain and predicting the shelf-life of temperature sensitive products inside the cold chain. In the study, real-time data of ambient parameters are gathered using IEEE 802.15.4 based wireless sensor networks and sent to the remote server through a gateway so that the shelf life of the products can be predicted by the decision support system developed. Radio Frequency Identification (RFID) is also used for identification of perishable goods inside the cold chain. All the devices and protocols employed in the study are modeled and simulated using event-driven Riverbed Modeler software.
Analyzing effects of random perturbation approaches on the popularity bias issue of recommendation algorithms and developing novel fake rating injection-based solutions
This thesis explores the advancement of recommender systems, with a focus on addressing popularity bias through privacy-preserved collaborative filtering techniques. Recommender systems are pivotal in filtering vast information spaces, guiding users towards items of interest. However, these systems often suffer from popularity bias, where popular items are disproportionately recommended, overshadowing less known items. This work introduces two novel approaches: a robust method for privacy-preserved collaborative filtering and the EquiRate algorithm for mitigating popularity bias. The privacy-preserved collaborative filtering technique employs randomized perturbation and obfuscation to safeguard user privacy while maintaining recommendation quality. This method not only enhances user trust by protecting sensitive information but also contributes to the accuracy and reliability of the recommendations. On the other hand, the EquiRate method specifically addresses the challenge of popularity bias. By integrating the FusionIndex metric, which assesses both recommendation accuracy and diversity, EquiRate efficiently balances the representation of popular and niche items, promoting a more diverse and equitable item exposure. Experimental evaluations on benchmark datasets reveal that these methods significantly improve recommendation diversity without compromising accuracy. The robust privacy-preserved collaborative filtering demonstrates resilience against various privacy attacks, ensuring effective recommendation under stringent privacy constraints. Meanwhile, EquiRate outperforms existing popularity-debiasing methods across multiple datasets, as evidenced by its superior FusionIndex scores. These outcomes highlight the potential of integrating privacy preservation and bias mitigation strategies to enhance the overall effectiveness of recommender systems. In conclusion, this thesis contributes to the recommender system literature by presenting innovative solutions for two pressing issues: privacy preservation and popularity bias. The proposed methods not only advance the state-of-the-art in collaborative filtering but also pave the way for creating more balanced, accurate, and privacy-conscious recommendation platforms.
Yapı izlenebilirlik çizgelerini kullanarak uzman geliştirici bulma
Mentoring is a commonly used practice in the software industry where mentors and mentees are matched to ease the onboarding process of the mentee, who is a newcomer. Also, during a project's life cycle, developers work on sections of the codebase that are unfamiliar to them. Both cases raise the task of finding an expert developer to contact for possible questions. With this study, we aim to construct an algorithm that recommends expert developers for a specific part of the codebase, namely folders, files, and methods, based on previous developer activities such as commits and code reviews. We construct an artifact traceability graph using commit history, method change history, code review history, and issue history. The relationships in the graph are weighted according to recency and a weight coefficient we determine intuitively. Utilizing this graph, we calculate a score representing the developer's expertise level on a folder, file, or method, and recommend developers with the highest expertise. To evaluate the success of our algorithm, Expert Developer Finder, we compare its recommendation with the developers who commented on related issues. We run our algorithm on three open-source projects - Nutch, OpenNLP, and Curator. On average, for weighted recommendations, we reached up to 84\% accuracy for folders, 82\% accuracy for files, and 88\% accuracy for methods. On average, for unweighted recommendations, we reached up to 84\% accuracy for folders, 84\% accuracy for files, and 93\% accuracy for methods. We believe that our results show that the Expert Developer Finder algorithm is able to recommend experts by utilizing the historical data of projects. However, further work is required to fine-tune the weights set in the artifact traceability graph.
Kalın bağırsak bezlerinin kurala dayanarak bölütlenmesi
Colon adenocarcinoma, which accounts for more than 90 percent of all colorectal cancers, originates from epithelial cells that form colon glands. Thus, for its diagnosis and grading, it is important to examine the distortions in the organizations of these epithelial cells, and hence, the deformations in the colon glands. Therefore, localization of the glands within a tissue and quantification of their deformations is essential to develop an automated or a semi-automated decision support system. With this motivation, this thesis proposes a new structural segmentation algorithm to detect glands in a histopathological tissue image. This structural algorithm proposes to transform the histopathological image into a new representation by locating a set of primitives using the Voronoi diagram, to generate gland candidates by defining a set of rules on this new representation, and to devise an iterative algorithm that selects a subset of these candidates based on their fitness scores. The main contribution of this thesis is the following: The representation introduced by this proposed algorithm enables us to better encode the colon glands by defining the rules and the fitness scores with respect to the appearance of the glands in a colon tissue. This representation and encoding have not been used by the previous studies. The experimental results of our algorithm show that this proposed algorithm improves the segmentation results of its pixel-based and structural counterparts without applying any further processing.
Transfer öğrenimi ile kestirimci bakım
Breakdown prediction of equipment is an essential task considering the management of resources and maintenance operations. Early diagnosis systems allow creating alerts on time for taking precautions on production. A significant challenge for diagnosis is to have an insufficient size of data, yet, transfer learning approaches can alleviate such an issue when there is a constrained supply of training data. We intend to improve the reliability of breakdown prediction when there is a limited quantity of training data. We recommend similarity correlation on Remaining Useful Life of these equipment. To do this, we offer learning a common feature space between the target and the source equipment, where we acquire prior knowledge from the source that has different measurements than the target. Within the learned joint feature matrices, we train our model on the vast amount of data of different equipment and finetune it using the data of our target equipment. In this way, we aim to obtain an accurate and reliable model for early breakdown prediction.
Olasılıksal gradyan alçalmanın hibrit paralelleştirilmesi
The purpose of this study is to investigate the efficient parallelization of the Stochastic Gradient Descent (SGD) algorithm for solving the matrix completion problem on a high-performance computing (HPC) platform in distributed memory setting. We propose a hybrid parallel decentralized SGD framework with asynchronous communication between processors to show the scalability of parallel SGD up to hundreds of processors. We utilize Message Passing Interface (MPI) for inter-node communication and POSIX threads for intra-node parallelism. We tested our method by using four different real-world benchmark datasets. Experimental results show that the proposed algorithm yields up to 6 times better throughput on relatively sparse datasets, and displays comparable performance to available state-of-the-art algorithms on relatively dense datasets while providing a flexible partitioning scheme and a highly scalable hybrid parallel architecture.
Dağıtık sistemlerde çok düzeyli görev atama algoritmaları
Task assignment problem deals with assigning tasks to processors in order to minimize the sum of execution and communication costs in a distributed sys tem. In this work, we propose a novel task clustering scheme which considers the differences between the execution times of tasks to be clustered as well as the communication costs between them. We use this clustering approach with proper assignment schemes to implement two-phase assignment algorithms which can be used to find suboptimal solutions to any task assignment prob lem. In addition, we adapt the multilevel scheme used in graph/hypergraph partitioning to the task assignment. Multilevel assignment algorithms reduce the size of the original problem by collapsing tasks, find an initial assignment on the smaller problem, and then projects it towards the original problem by successively refining the assignment at each level. We propose several clus tering schemes for multilevel assignment algorithms. The performance of all proposed algorithms are evaluated through an experimental study where the as signment qualities are compared with two up-to-date heuristics. Experimental results show that our algorithms substantially outperform both of the existing heuristics. Key words: Task assignment, distributed systems, task clustering, multilevel task assignment methods, Kernighan-Lin Heuristic.
Bulut bilişim sistemlerinde sanal makineler üzerinde taşıyıcılar ile yığın iş çizelgelemesi
Virtualization and use of virtual machines (VMs) is important for both public and private cloud systems and also for users. The allocation and use of virtual machines can be optimized by using knowledge about expectations of users, such as resource demands, network communication patterns, and total budget. However, both public and private cloud providers do not expose advanced configuration options to make use of custom needs of users. Adding upon to previous research, we propose a new approach for allocating and scheduling user jobs to virtual machines by use of container technologies like Docker, so that VM utilization can be increased and costs for users can be decreased. In our approach, by predicting resource demands, we can schedule different kinds of jobs on a single virtual machine without jobs affecting each other and without degrading performance to unacceptable levels. We also allow cost-performance tradeoff for users. We verified our approach in a real test-bed and evaluated it with extensive simulation experiments. We also adapted our approach into a real web-based application we developed, called PAGS (Programming Assignment Grading System), which enables efficient and convenient testing, submission and evaluation of programming assignments of a large number students in an interactive or batch manner in identical and isolated system environments. Our approach effectively schedules requests from teachers and students so that the system can horizontally scale in a cost efficient manner.
Veri birleştirme ve Wikipedia kullanarak küme etiketlemenin iyileştirilmesi
A cluster is a set of related documents. Cluster labeling is the process of assigning descriptive labels to clusters. This study investigates several cluster labeling approaches and presents novel methods. The first uses clusters themselves and extracts important terms, which distinguish clusters from each other, with different statistical feature selection methods. Then it applies different data fusion methods for combining their outcomes. Our results show that although it provides statistically significantly better results for some cases, it is not a stable and reliable labeling method. This can be explained by the fact that a good label may not occur in the cluster at all. The second exploits Wikipedia as an external resource and uses its anchor texts and categories to enrich the label pool. Labeling with Wikipedia anchor text fails because the suggested labels tend to focus on minor topics. Although the minor topics are related to the main topic, they do not exactly describe it. After this observation, we use categories of Wikipedia pages to improve our label pool in two ways. The first fuses important terms and Wikipedia categories with rank based fusion methods. The second looks relatedness of Wikipedia pages to the clusters and use only categories of related pages. The experimental results show that both methods provide statistically significantly better results than the other cluster labeling approaches that we examine in this study.
Wikipedia yolu ile bağlamsal ilişki filtrelemesi kullanarak geliştirilmiş özellik seçme
Feature selection is an important component of information retrieval and natural language processing applications. It is used to extract distinguishing terms for a group of documents; such terms, for example, can be used for clustering, multi-document summarization and classification. The selected features are not always the best representatives of the documents due to some noisy terms. Addressing this issue, our contribution is twofold. First, we present a novel approach of filtering out the noisy, unrelated terms from the feature lists with the usage of contextual relatedness information of terms to their topics in order to enhance the feature set quality. Second, we propose a new method to assess the contextual relatedness of terms to the topic of their documents. Our approach automatically decides the contextual relatedness of a term to the topic of a set of documents using co-occurrences with the distinguishing terms of the document set inside an external knowledge source, Wikipedia for our work. Deletion of unrelated terms from the feature lists gives a better, more related set of features. We evaluate our approach for cluster labeling problem where feature sets for clusters can be used as label candidates. We work on commonly used 20NG and ODP datasets for the cluster labeling problem, finding that it successfully detects relevancy information of terms to topics, and filtering out irrelevant label candidates results in significantly improved cluster labeling quality.
Nitel maliyete duyarlı sınıflandırma
Decision making is a procedure for selecting the best action among severalalternatives. In many real-world problems, decision has to be taken under thecircumstances in which one has to pay to acquire information. In this thesis, wepropose a new framework for test-cost sensitive classification that considers themisclassification cost together with the cost of feature extraction, which arisesfrom the effort of acquiring features. This proposed framework introduces twonew concepts to test-cost sensitive learning for better modeling the real-worldproblems: qualitativeness and consistency.First, this framework introduces the incorporation of qualitative costs intothe problem formulation. This incorporation becomes important for many realworld problems, from finance to medical diagnosis, since the relation betweenthe misclassification cost and the cost of feature extraction could be expressedonly roughly and typically in terms of ordinal relations for these problems. Forexample, in cancer diagnosis, it could be expressed that the cost of misdiagnosisis larger than the cost of a medical test. However, in the test-cost sensitive classificationliterature, the misclassification cost and the cost of feature extractionare combined quantitatively to obtain a single loss/utility value, which requiresexpressing the relation between these costs as a precise quantitative number.Second, the proposed framework considers the consistency between the currentinformation and the information after feature extraction to decide which featuresto extract. For example, it does not extract a new feature if it brings no newinformation but just confirms the current one; in other words, if the new featureis totally consistent with the current information. By doing so, the proposedframework could significantly decrease the cost of feature extraction, and hence,the overall cost without decreasing the classification accuracy. Such consistencybehavior has not been considered in the previous test-cost sensitive literature.We conduct our experiments on three medical data sets and the results demonstratethat the proposed framework significantly decreases the feature extractioncost without decreasing the classification accuracy.
Sanal şehir modelleme ve nüfuslandırma: Otomatik bina modeli üretimi ve acil durumlar icin kalabalık simülasyonu
In this thesis, we present an automatic building generation method based on procedural modeling approach, and a crowd animation system that simulates a crowd of pedestrians inside a city. While modeling the buildings, to achieve complex and consistent geometries we use shape grammars. The derivation process incorporates randomness so the produced models have the desired variation. The end shapes of the building models could be defined in a certain extent by the derivation rules. The behavior of human crowds inside a city is affected by the simulation scenario. In this thesis, we specifically intend to simulate the virtual crowds in emergency situations caused by an incident, such as a fire, an explosion, or a terrorist attack. We prefer to use a continuum dynamics-based approach to simulate the escaping crowd, which produces more efficient simulations than the agent-based approaches. Only the close proximity of the incident region, which includes the crowd affected by the incident, is simulated. In order to speed up the animation and visualization of the resulting simulation, we employ an offline occlusion culling technique. During runtime, we animate and render a pedestrian model only if it is visible to the user. In the pre-processing stage, the navigable area of the scene is decomposed into a grid of cells and the from-region visibility of these cells is computed with the help of hardware occlusion queries.
İçerik tabanlı görüntü erişimi için anlamsal sahne sınıflandırması
Content-based image indexing and retrieval have become important researchproblems with the use of large databases in a wide range of areas. Because of theconstantly increasing complexity of the image content, low-level features are nolonger sufficient for image content representation. In this study, a content-basedimage retrieval framework that is based on scene classification for image indexingis proposed. First, the images are segmented into regions by using their color andline structure information. By using the line structures of the images the regionsthat do not consist of uniform colors such as man made structures are captured.After all regions are clustered, each image is represented with the histogram ofthe region types it contains. Both multi-class and one-class classification modelsare used with these histograms to obtain the probability of observing differentsemantic classes in each image. Since a single class with the highest probabilityis not sufficient to model image content in an unconstrained data set with a largenumber of semantically overlapping classes, the obtained probability values areused as a new representation of the images and retrieval is performed on thesenew representations. In order to minimize the semantic gap, a relevance feedbackapproach that is based on the support vector data description is also incorporated.Experiments are performed on both Corel and TRECVID datasets and successfulresults are obtained.
Doküman numaralarını yeniden atama yolu ile ters indeks sıkıştırma
Compression of inverted indexes received great attention in recent years. An inverted index consists of lists of document identifiers, also referred as posting lists, for each term. Compressing an inverted index reduces the size of the index, which also improves the query performance due to the reduction on disk access times.In recent studies, it is shown that reassigning document identifiers has great effect in compression of an inverted index. In this work, we propose a noveltechnique that reassigns both term and document identifiers of an inverted index by transforming the matrix representation of the index into a block-diagonal form, which improves the compression ratio dramatically. We adapted row-net hypergraph-partitioning model for the transformation into block-diagonal form, which improves the compression ratio by as much as 50%. To the best of our knowledge, this method performs more effectively than previous inverted index compression techniques.
Kelime eşlenmesi içim çizgi tabanlı bir niteleme
With the increase of the number of documents available in the digital environment, efficient access to the documents becomes crucial. Manual indexing of the documents is costly; however, and can be carried out only in limited amounts. Therefore, automatic analysis of documents is crucial. Although plenty of effort has been spent on optical character recognition (OCR), most of the existing OCR systems fail to address the challenge of recognizing characters in historical documents on account of the poor quality of old documents, the high level of noise factors, and the variety of scripts. More importantly, OCR systems are usually language dependent and not available for all languages. Word spotting techniques have been recently proposed to access the historical documents with the idea that humans read whole words at a time. In these studies the words rather than the characters are considered as the basic units. Due to the poor quality of historical documents, the representation and matching of words continue to be challenging problems for word spotting. In this study we address these challenges and propose a simple but effective method for the representation of word images by a set of line descriptors. Then, two different matching criteria making use of the line-based representation are proposed. We apply our methods on the word spotting and redif extraction tasks. The proposed line-based representation does not require any specific pre-processing steps, and is applicable to different languages and scripts. In word spotting task, our results provide higher scores than the existing word spotting studies in terms of retrieval and recognition performances. In the redif extraction task, we obtain promising results providing a motivation for further and advanced studies on Ottoman literary texts.
Arduıno ve bluetooth kullanılarak uzaktan kontrollü robot tasarım, yazılım ve uygulamasının yapılması
Robotik uygulamalar teknolojinin ilerlemesiyle birlikte günlük hayatta ve endüstriyel otomasyonlarda daha yaygın olarak kullanılmaya başlamıştır. Bu sistemler güvenirlilik ve hız açısından hem daha doğru sonuçlar vermekte hem de bu sonuçlara daha ekonomik şekilde ulaşılabilmektedir. Robotik uygulamalarda genellikle gezginlik kabiliyeti önem kazanır. Bizlerin bugün bile basit olarak tanımlayabileceğimiz gezgin robotları önemli kılan şey, şu an yaptıkları ve insanoğlunun yaratıcılığını kullanarak ilerde yapabilecekleri işlerdir. Bu projemiz, bir araç platformu kullanılarak uzaktan kontrollü elektromekanik bir sistemde, android tabanlı bir telefon ile bluetooth üzerinden bir aracın kontrol edilmesidir. Teknolojinin ilerlemesiyle birlikte insanların yaşamlarını kolaylaştıran cihazlara yönelmesi ve bu cihazların tek elden kontrol edilmesi istendiği gözlemlenmiştir. Bu kontrol sırasında gereksiz kablo ve kullanımı zor olan aletlerden kaçınılmaktadır. Bu nedenle günümüzde yavaş yavaş bütün cihazların kablosuz cihazlar üzerinden kontrolüne geçiş yapılmaktadır.Bu projede Arduino UNO R3, HC05 bluetooth modülü, L293 motor shield, 12V kuru akü, DC motor ve pleksiglass gövde kullanılmıştır. Projenin yazılım kısmı ise gömülü yazılım Arduino'nun arayüzü ile, android yazılımı ise App Inventor programı ile yazılmıştır. Android cihazdan gönderilen veriler, HC05 bluetooth modülü üzerinden Arduino UNO'ya gönderilmektedir. Arduino ise gelen verilerin gerekli şartları sağlaması halinde L293 Motor Shield'i tetikleyerek motorların gereken yön ve hızda dönmesini sağlamaktadır.
Sezgisel algoritmalarla kesir dereceli pıda denetçi tasarımı ve bozucu dışlama performansının iyileştirilmesi
Sistemlerin kontrol performansının daha da iyileştirilmesi amacıyla uzun yıllardan beri çeşitli denetçi tasarım modelleri geliştirilmektedir. Özellikle hassas kontrol uygulamalarında geliştirilen bu denetçi modellerinin bozucu dışlama performansının da iyileştirilmesi önemli bir araştırma konusu olmuştur. Bu tez çalışmasında, literatürde yüksek dereceli sistemlerin kontrolünde etkili olduğu öne sürülen PIDA denetçilerin tasarımı ve bozucu dışlama performansının iyileştirilmesi konusunda çalışmalar yapılmıştır. Bu amaçla bu denetçilerin önce tamsayı dereceli daha sonra da kesir dereceli versiyonları ele alınmıştır. Bu kapsamda Bölüm 3'te önce PIDA denetçi parametrelerinin PSO algoritması ile belirlenmesi üzerine bir yöntem geliştirilmiş, bu tasarım yöntemi ile elde edilen PIDA denetleyicinin başarımı, literatürdeki Gradiyant arama ve Genetik algoritma optimizasyon yöntemleri ile karşılaştırılmış, simülasyon sonuçlarının daha iyi performans sergilediği görülmüştür. Ayrıca, optimum PIDA denetleyici, SOS algoritması ile tasarlanmış ve kontrol performansı, mevcut arama algoritmalarından Gradiyant Arama ile karşılaştırılmış bozucu dışlama performansının daha iyi olduğu gösterilmiştir. Daha sonra PIA denetçi parametreleri, RDR ölçütü ile kontrol hatası arasında bir uzlaşma sağlanarak RA optimizasyon algoritması ile tasarlanmış ve birim basamak yerleşme noktası kontrol performansı ile bozucu dışlama performanslarının birlikte iyileştirilebildiği gösterilmiştir. Diğer bir çalışmada da RA algoritması çoklu amaç fonksiyonu ile kullanılarak, birim basamak referans giriş kontrol performansı ile bozucu dışlama performanslarının birlikte iyileştirilebildiği, referans giriş filtreli iki serbestlik dereceli kapalı çevrim PIDA denetçi RDR analizine dayalı olarak tasarlanmıştır. Bölüm 4'te ise önceki bölümde önerilen yöntemler, denetçilerin kesir dereceli versiyonlarına uyarlanmıştır. Bu kapsamda, önce iki serbestlik dereceli bir sistem için KDPID tasarımı yapılmış ve bozucu dışlama performansı iyileştirilmiştir. Daha sonra, referans giriş filtresi içeren bir 2DOF KDPIDA kontrol sistemi tasarımı için uzlaşma eğrisi odaklı UORA algoritması önerilerek bu algoritma ile RDR spektrumu klavuzunda, KDPIDA denetçinin hem referans giriş kontrol performansının hem de bozucu dışlama performansının birlikte iyileştirilebildiği bir tasarım tekniği önerilmiştir. Anahtar Kelimeler: Kesir Dereceli PIDA Denetçi, Optimizasyon, Bozucu dışlama
Düzensiz hacimsel ağların donanımsal hızlandırıcı yöntemleri ile doğrudan görüntülenmesi
Computational fluid dynamic simulations often produce large clusters of finite elements with non-trivial, non-convex boundaries and uneven distributions among compute nodes, posing challenges to compositing during interactive volume rendering. Correct, in-place visualization of such clusters becomes difficult because viewing rays straddle domain boundaries across multiple compute nodes. We propose a GPU-based, scalable, memory-efficient direct volume visualization framework suitable for in situ and post hoc usage. Our approach reduces memory usage of the unstructured volume elements by leveraging an exclusive or-based index reduction scheme and provides fast ray-marching-based traversal without requiring large external data structures built over the elements. Moreover, we present a GPU-optimized deep compositing scheme that allows correct order compositing of intermediate color values accumulated across different ranks that works even for non-convex clusters. Furthermore, we illustrate that we can achieve secondary effects such as shadows and gradient shading using our method for single GPU setups. Our approach scales well on large data-parallel systems and achieves interactive frame rates during visualization. We can interactively render Fun3D Small Mars Lander (14 GB / 798.4 million finite elements) and Huge Mars Lander (111.57 GB / 6.4 billion finite elements) data sets at 14 and 10 frames per second using 72 and 80 GPUs, respectively, on the Frontera supercomputer at The Texas Advanced Computing Center (TACC).
Gaz bazlı araçlar için akıllı hırsızlığa karşı güvenlik ve güvenlik sistemi tasarımı ve uygulaması
In the past few years, many cases of car theft operations have been recorded to utilised it in suspicious operations. In addition to some countries 'lack of oil, these countries tended to modify cars to run on liquid propane gas. Consequently, we have many problems related to protecting cars from theft and others to protect gas-based cars, such as temperature rising or gas leaks, which subsequently cause dangerous accidents. In this work, intelligent system is proposed to protect the cars from the theft and the accidents related to gas-based cars. The proposed system comprises of three major portions; the first one is the car security subsystem, the following part is the real time car tracking subsystem, and the last part is the car safety subsystem. The car security subsystem is based on the Arduino Microcontroller, Bluetooth module, vibration sensor, keypad, solenoid lock, and GSM module. When the car owner would operate his own car a special password should be entered from the keypad and another password from a special Android app, once the two passwords right the car will be operated normally. If one of the entered passwords is error more than three times or the car operated suddenly without entering the passwords the alarming system will be turned ON, the doors will be closed by the solenoid lock, and a warning message will be sent to the car owner via the Bluetooth module. The car tracking subsystem is based on the internet of things () technology with a programmed android app utilising the NodeMCU microcontroller and global positioning system. Finally, the car safety system is based on the Arduino UNO microcontroller board, MQ-2 gas sensor to reveal the gas leak, flame sensor to reveal the appearance of the fire, the temperature sensor to monitor the ambient temperature of a vehicle's gas container, and Bluetooth module. In addition, the user will be able to monitor all the status of the safety (i.e. temperature, gas leak, and flame detection) via the compatible android app
Designing a smart system to detect the intrusion in IoT
The Internet of Things (IoT) has been quickly growing during the previous few years, with the intention of having a wider impact on every aspect of life, from daily activities to vast industrial systems. Unluckily, A group of cybercriminals took notice of this, those responsible for turning the IoT into a vector for cybercrime, potentially exposing end nodes to attack. Since there are so many different kinds of IoT devices, There are difficulties to defend infrastructure for the IoT using a normal intrusion detection system (IDS). IoT devices need to be protected. We looked at data flow in the IoT. in this work taking two datasets (UNSW-NB15 and DoH20) and using three Machine Learning (ML) classifiers: Random Forest (RF), K-Nearest Neighbors (KNN) and Decision Tree (DT). For each method, we determined the Error Rate (ER), Accuracy(Acc), Precision, Recall, and F1 score. We received outstanding results (100%) when we combined these two classifiers. Detection rates are extremely high. A succinct summary of the results is presented.
Nesnelerin interneti kullanılarak gelişmiş akıllı otopark sistemi geliştirildi
Automobile use has increased dramatically in recent years as a result of the automotive industry's rapid growth and the daily release of more convenient and technologically advanced vehicles to the market. On the other hand, the increasing usage of autos creates additional parking challenges. Throughout the preceding decades, a huge number of researchers have worked on this project. They were striving to get intelligent technology that would assist them in resolving the parking dilemma, and they had devised a system to do so. The Internet of Things (IoT) idea was applied to the development of a smart parking system for autos in this research (IoT). Wireless sensors and the network layer were used in this system to find the vehicle's exam-free parking space, which was determined via the network layer. C++ was chosen as the programming language for the system's software component, which was developed by many researchers. Along with the coding portion of the article, an output window was included. When you come together with other people, you can have a meaningful conversation regarding smart parking systems and the Internet of Things.
Naive bayes algoritmasını kullanarak kötü amaçlı yazılım url'sini algılama
Hacking and fake pages are the basis of problems and suspicious activities on the Internet, Therefore, the disadvantages of those pages are the reason for the increased request for safeguard which prevents the user from accessing, our study explains the possibility of identifying suspicious links from the URL-based features of its addresses, we demonstrate that our problem is consistent with machine learning algorithms, it also fits to the modern features of the continuously evolving distribution of malicious URLs, we have also developed the model for those predictive addresses and categorized it into safe or unsafe URL by using naive Bayes algorithm we also compare this work with the researchers' other studies.
Akıllı ve güvenli ev için şeylerin interneti tabanlı zigbee snıffer
This thesis aims to resolve the Internet of Things (IoT) based ZigBee sniffer for smart home and determine the usage of energy or power with high spectrum allocation in future ZigBee Protocol with the help of clustering in IoT with data mining. The research work starts presenting an overview of the broadband network energy sector and the challenges that face it. It is observed a change in the energy policies promoting energy efficiency, encouraging an active role of the consumer, instructing them about the importance of consumer behavior, and protecting consumer rights. Electricity is gaining room as an energy source. Its share will keep constantly increasing in the following decades. ZigBee Protocol and smart meters' deployment will benefit both the utility and the consumer in the near future. New services and new businesses appear in this environment, focusing on the energy management field and tools. They require specialization in fields such as computer science, software development, and data science. This research has segmented the ZigBee Protocol according to the similarities of their electrical load profiles, using the proportion of energy usage per hour (%) as a common framework. This energy consumption segmentation aims to provide personalized recommendations to each group to reduce their energy consumption and the associated costs, fostering energy efficiency measures and improving consumer engagement. The desired segmentation is obtained by an iterative process, based on computational clusters calculation (using a Python programming language) and finalized by a post-clustering analysis applying visualization and statistical data mining technique to detect the energy consumption and reallocate them to a more appropriate group. The K-Means clustering technique was tested and compared, giving the best prediction of accuracy 98.46% for all energy load profiles with a high spectrum of 100GHz. The solution from the K-Means clustering is the one that better adapts to the segmentation sought, which is used as the base of the post-clustering stage to obtain the final energy consumption segmentation. Most of these methodologies use the absolute values in 100 kWh, focusing on identifying the users with higher energy savings potential. The energy consumption segmentation of the electricity consumers provides knowledge and a better understanding of the consumer. In this particular case, it allows to personalize energy savings recommendations according to the ZigBee protocol-specific characteristics, improving the consumer experience by providing adequate advice at the appropriate time, facts that increase the effectiveness of the energy efficiency advice' service future ZigBee protocol. Keywords: Internet of Things (IoT), K-Means Clustering, Energy Consumption, Smart, Zigbee protocol
Genetik algoritmaların çaprazlama, mutasyon metodlarının ve parametrelerinin gezgin satıcı problemi üzerinde analizi
With the rapid development of whole industry(automotive and especially logistics) and software industry, increasing demand by customers and supply by manifacturers led the optimization more and more important nowadays. By the word for optimization, we mean minimizing production times, maximizing product logictics per transportation or minimizing fuel usage/maximizing fuel saving/efficiency for transportation vehicles. By the demand of these optimizations by the industry, also led optimization algorithms/techniques to grow and evolve. With the evolution of computers and computation powers, classic optimization techniques also evolved. One of evolutinary optimization techniques, Genetic algorithms and genetic programming, corresponded to these heavy demand of optimization area. Basically, genetic algorithms evolved from genetics and applications of sir Charles Darwin, crossover and mutation principles. Using Genetic Algorithms, we have the ability to optimize our solutions for hard problems. Simply, finding/choosing random solutions to the problem and make crossover and mutations on these solutions as the nature does. Crossing over and mutate the parts of solutions by switching the meaningful data between solutions and hope to reach to the best optimized solution. Generally we reach to the optimized solution by finding and trying correct or better crossover and mutation rates. In other words, choosing bad rates for these parameters, most likely leads to worse optimization. In this work, firstly, we presented the genetic algorithms in general way and after that we go in deep and used genetic algorithms to find better optimized results for the famous Traveling Salesman Problem. We chose to apply genetic algorithms on geographical regions of Turkey(Marmara, Aegean and Black Sea regions, 32 cities in total) to find best or best optimized route to travel. While applying genetic algorithms, we modified crossover methods, mutation methods and crossover and mutation rates to reach to the best possible route and analysed final solutions for each used parameter/method and made a comparison between them. Finally, we presented the compared results on graphics to visualize the evolution for each presented parameter. By making these research, we aim to reach out the best or better parameters for real use cases used in the logistics industry to reach better fuel efficiency and reducing fuel costs.
Çevrimci eğitim için kavram önkoşul haritalarının oluşturulması ve görselleştirilmesi
The growth of internet technologies in the last decade allowed the knowledge to spread out very fast across the globe. Users began educating themselves with huge amounts of online material on the internet. Today many academic institutions offer publicly available courses where students, all around the world can join and benefit from them. The abundance of online material from many different resources created an unorganized content in which it is likely for the learners to get lost. Furthermore, some concepts may require knowledge from other concepts and the learner may not be aware of those prerequisite relations between the concepts, therefore, he or she may have difficulties in understanding them. In our work, we create a metric for calculating a prerequisite score between two text-based educational material. We choose Wikipedia articles to work with since it is a large encyclopedia containing huge amounts of information on lots of different concepts. Furthermore, from a given set of concepts with their corresponding Wikipedia articles, we calculate each concept's prerequisite score with the other concepts and build a prerequisite concept graph for the learner. We hope that our graph model will guide the students in their studies and enhance their learning experience.
Şifrelenmiş veri üzerinde tümüyle güvenli, uygulanabilir, derecelendirilmiş ve çoklu anahtar kelime destekleyen arama methodu
Cloud computing offers computing services such as data storage and computing power and relieves its users of the burden of their direct management. While being extremely convenient, therefore immensely popular, cloud computing instigates concerns of privacy of outsourced data, for which conventional encryption is hardly a solution as the data is meant to be accessed, used and processed in an efficient manner. Multi keyword ranked search over encrypted data (MRSE) is a special form of secure searchable encryption (SSE), which lets users to privately find out the most similar documents to a given query using document representation methods such as tf-idf vectors and metrics such as cosine similarity. In this work, we propose a secure MRSE scheme that makes use of both a new secure k-NN algorithm and somewhat homomorphic encryption (SWHE). The scheme provides data, query and search pattern privacy and is amenable to access pattern privacy. We provide a formal security analysis of the secure k-NN algorithm and rely on IND-CPA security of the SWHE scheme to meet the strong privacy claims. The scheme provides speedup of about two orders of magnitude over the privacy-preserving MRSE schemes using only SWHE while its overall performance is comparable to other schemes in the literature with weaker forms of privacy claims. We present implementations results including one from the literature pertaining to response times, storage and bandwidth requirements and show that the scheme facilitates a lightweight client implementation.
Hasta kümelemesi için yolak çizge çekirdeği bazlı bir çoklu-omik yaklaşımı
Accurate classification of patients into molecular subgroups is critical for the development of effective therapeutics and for deciphering the underlining mechanisms for these subgroups. The availability of multi-omics data catalogs for large cohorts of cancer patients provides multiple views into the molecular biology of the tumors and the alterations that take place in patient genes such as mutations and differential expression patterns. At the same time, the molecular interaction networks provide the biological context for these alterations. We develop PAMOGK (Pathway based Multi Omic Graph Kernel clustering) that integrates multi-omics patient data with existing biological knowledge on pathways. We use a novel graph kernel that evaluates patient similarities based on a single molecular alteration type in the context of a pathway. To corroborate multiple views of patients that are evaluated by hundreds of pathways and molecular alteration combinations, we use a multi-view kernel clustering approach. Applying PAMOGK to kidney renal clear cell carcinoma (KIRC) patients results in four clusters with significantly different survival times (p-value = 1.24e-11). When we compare PAMOGK to eight other state-of-the-art multi-omics clustering methods, PAMOGK consistently outperforms these in terms of its ability to partition KIRC patients into groups with different survival distributions. The discovered patient subgroups also differ with respect to other clinical parameters such as tumor stage and grade, and primary tumor and metastasis tumor spreads. The pathways identified as important are highly relevant to KIRC. We also extend our analysis to eight other cancer types with available mutation, protein, and gene expression data. PAMOGK framework is available in https://github.com/tastanlab/pamogk
Yarı denetimli derin kümeleme yaklaşımıyla çapraz kanser hastalarının belirlenmesi
In traditional medicine, the treatment decisions for a cancer patient are typically based on the patient's cancer type. The availability of molecular profiles for a large cohort of multiple cancer patients opens up possibilities to characterize patients at the molecular level. There have been reports of cases where patients with different cancers bear similarities. Motivated from these observations, in this thesis, we specifically focus on developing a method to discover cross-cancer patients. We define cross-cancer patients as those who have molecular profiles that bear a high level of similarity to other patient(s) diagnosed with a different cancer type and are not representative of their cancer type. To find cross-cancer similar patients, we develop a framework where we identify patients that co-cluster frequently when clustered based on their transcriptomic profiles. To solve the clustering problem, we propose a semi-supervised deep learning clustering in which the clustering task is guided by the cancer types of the patients and the survival times. The deep representation obtained in the network is used in the clustering module of DeepCrossCancer. Applying the method to nine different cancers from The Cancer Genome Atlas project using patient tumor gene expression data, we discover twenty patients similar to a patient or multiple patients in another cancer type. We analyze these patients in light of other genomic alterations. Our results find significant similarities both in mutation and copy number variations of the cross-cancer patients. The detection of cross-cancer patients opens up possibilities for transferring clinical decisions from one patient to another and expediting the investigation of novel cancer drivers shared among them.
Örtük otokodlayıcı ve wasserstein kaybı içeren döngüsel çekişmeli çerçeve (CAFIAWL)
Since the day that the Simple Perceptron was invented, Artificial Neural Networks (ANNs) attracted many researchers. Technological improvements in computers and the internet paved the way for unseen computational power and an immense amount of data that boosted the interest (therefore the advance), particularly in the last decade. As of today, NNs seem to take a vital role in all different types of machine learning research and the main engine of many applications. Not only learning from the data with machines in order to make informed decisions but also "creating" something new, unseen, novel with machines is also a very appealing area of research. The generative models are among the most promising models that can address this goal and eventually lead to "computational creativity". Recently the Variational Autoencoders (VAE) and the Generative Adversarial Networks (GAN) have shown tremendous success in terms of their generative performance. However, the conventional forms of VAEs had problems in terms of the quality of the outputs and GANs suffered hard from a problem that limited the diversity of the generated outputs, i.e., the mode collapse problem. One line of research that targets to eliminate these weaknesses of both algorithms is developing hybrid models which capture the strengths of these algorithms but avoiding their weaknesses. In this research, we propose a novel generative model. The proposed model is composed of four adversarial networks. Two of the adversarial networks are very similar to conventional GANs and the remaining two are basically WGAN that is based on the Wasserstein loss function. The way that these adversarial networks are put together also incorporates two implicit autoencoders to the proposed model and provides a cyclic framework that addresses the mode collapse problem. The performance of the proposed model is evaluated in various aspects by using the MNIST data. The analysis suggests that the proposed model generates good quality output meanwhile avoids the mode collapse problem.
Kombinatoryal etkileşim test tabanlı günlük derleme süreci
A daily build process is a process where the latest version of a software under development is obtained from its code repository on a daily basis (typically during off-work hours), configured, built, and tested against a test suite. The ultimate goal of this process is to reveal defects in the most fundamental functionalities of the system as soon as they are introduced into the codebase, so that the turnaround time for fixing them is reduced as much as possible. In this work, we first introduce combinatorial interaction testing-based daily build process where a combinatorial object is computed to systematically test the interactions between system parameters on a daily basis. We then introduce a number of different testing strategies and empirically demonstrate that the proposed approach profoundly improves the effectiveness of the standard daily build processes.
CPU-GPU heterojen veritabanı yönetim sisteminde makine öğrenmesi kullanarak iş dağıtımı
Conventional OLTP systems are slow in performance for analytical queries. In the existing heterogeneous architecture OLAP database management systems, no system distributes work using machine learning. In this study, the DOLAP architecture, which is a high-performance column-based database management system developed for shared memory architectures, is explained. Also, job distribution algorithms based on heuristic and machine learning methods have been developed for computing hardware with different characters such as CPU and GPU on the server on which the database is running, and their performance has been analyzed.
Android uygulamalarda sistematik örnekleme ile otomatikleştirilmiş model keşif yaklaşımı
Clients progressively depend on mobile applications for computational needs. With the popularity of Google Android and the rise of interest in Android devices, Android applications have been valuable and millions of mobile applications have increased the importance and demand of test processes in the complex systems. Since the applications had well-developed strong conditions that need to be tested, automation in the testing has played a significant role. Many types of researches have primarily focused on different model discovery strategies to be used for different purposes (e.g., test generation, bug detection). However, they were not used systematically for testing of mobile applications. We present a tool that provides an automated black-box model discovery by applying systematic sampling to build a model of an application dynamically for different uses. The approach includes two purposes: (1) discovering the model of an application by providing systematic sampling, and (2) predicting guard conditions of the discovered model. The results of our experiments have confirmed the ability of the approach to acquire higher code coverage and the accuracy of predicted guard conditions than existing approaches.
Sonlu durum makinelerinde W-kümesi ve K-ağacı türetimi için sezgisel algoritmaların iyileştirilmesi
Finite State Machine (FSM) based testing methods utilize State Identification Sequences which are used to identify the states of a black box implementation as corresponding to states of an FSM given as the specification. There are different types of state identification sequences. Some of these state identification sequences are not guaranteed to exist for all specifications. There is one particular type of state identification sequences, W-set based state identification sequences, which are known to exist for any minimal, deterministic, completely specified FSM. Although W-set based state identification sequences are known for a very long time, most of the works in FSM based testing literature do not prefer to use them when testing for an implementation without a reliable reset feature, since the length of W-set based state identifications sequences in this case are exponential in the cardinality of the W-sets. There are some recent works that suggest to reduce the length of the W-set based state identification sequences. In fact, instead of W-sets, which are sets of preset experiments, these new methods can make use of so called K-sets, which are set of adaptive experiments that again always exist. Furthermore, these new methods suggest to apply not all elements of W–sets/K-sets, but instead an adaptive structure, called a K-tree is used to orchestrate the application of the elements of the K-set. However, there are no extensive experimental studies for these new methods. In addition, no algorithms are given for the construction of K-trees. In this work, we first present some W-set construction algorithms to construct better W-sets, in terms of both the cardinality and the total length of the sequences. We compare our W-set algorithms experimentally to the algorithms that exist in the literature. We also present algorithms to constructs K-sets and K-trees. Finally, we present an extensive experimental study for state identification sequences. The results show that, although W-set based state identification sequences have been considered practically infeasible due to the exponentially long sequences, the usage of K-trees make state identification sequences very short and practically usable. Utilizing K–sets in the generation of K–trees also yields better results than utilizing W–sets.
Multi-target detection & tracking using machine learning methodologies
This thesis presents a comprehensive study on multi-target human detection and tracking using machine learning methodologies, with a focus on integrating data from multiple camera sources. As the demand for advanced surveillance systems and automated monitoring increases, effective detection and tracking of individuals in dynamic environments become paramount. This research explores the implementation of state-of-the-art deep learning models, such as YOLO (You Only Look Once) and Faster R-CNN, for real-time human detection across varied environments. Building on existing methodologies, we introduce an innovative framework that not only detects and tracks primary individuals but also identifies and categorizes newly detected unknown individuals as "sub persons of interest." This hierarchical approach allows for enhanced relationship management between individuals and improves the accuracy of identity retention over time, especially in crowded or occluded scenarios. The proposed system employs advanced data association techniques, such as the Hungarian algorithm and Deep SORT, to seamlessly link detections across frames from multiple cameras while managing identity switches effectively. Evaluations conducted on diverse datasets highlight the effectiveness of the proposed method, demonstrating its robustness and scalability in real-world applications. Our findings indicate that leveraging multi-camera inputs significantly enhances detection and tracking performance, providing improved situational awareness. This research contributes to the growing field of computer vision by addressing current limitations in human detection and tracking systems, paving the way for future advancements in surveillance technologies and intelligent monitoring solutions. This work lays foundational concepts for future research into behavioral analysis and interaction recognition, further contextualizing the relationships between individuals in complex environments. Keywords: Multi-target tracking, Machine learning, Surveillance system
Protein etkileşimlerindeki sıcak noktaların analizi
Name of the Student : Sabri Bora ErdemliThesis Title : Analysis of Hot Spots in Protein - Protein InteractionsAbstractProteins generally function through reciprocal cooperation; they bind together intoprotein complexes and help each other fulfill their functions. Almost every level of cell functionis established by protein-protein interactions including formation of structure in cellularorganelles, the transport entities across the various biological membranes and signal transduction.The objective of this thesis is to investigate the role of energetically important residues called?hot spots? in protein associations. For this purpose 14 different complexes belonging to 4different complex types are analyzed. Molecular dynamics simulations are performed over theseprotein complexes. Hot spots and non hot spot residues in the interfaces are characterized withrespect to their flexibility and hydrogen bond forming ability. Interfaces are dominated byhydrophobic residues in almost all types of complexes. Then the second contribution is fromresidues that have aromatic side chains. Tyr, Ser, Phe, Gly and Thr are more frequent as hot spotresidues than the other residues in the interfaces. Hot spots are found to be more buried than therest of the interface. It is observed that the hot spot residues exhibit less mobility than the rest ofthe interface residues. The hydrogen bond formation ability of the hot spots is not significantlydifferent from the others. These findings should be very important in understanding protein -protein interaction and development of docking algorithms.
Türbülanslı yanma modelleri oyf denklemleri için yeni bir hibrit sonlu-hacimler/parçacık metodu ve hız modellerinin performans değerlendirilmesi
The consistent hybrid solution method has been recently developed and shown to be superior tothe best alternative approach by as much as a factor of 50 or more, making the PDF methodologya feasible design tool in practical applications (Muradoglu et al,2003). In the present study, thehybrid solution method is improved significantly by replacing the density-based finite-volume(FV) solver with a SIMPLE type pressure-based FV solver (Peric,1996) and the new solutionalgorithm is used to simulate the bluff-body stabilized turbulent non-reacting and reacting flowsstudied experimentally by Dally et al. (1999). Non-reacting and reacting bluff-body flowsimulations are performed by Jenny et al. (2001) and Muradoglu et al. (2003), respectively. Thevelocity-turbulent frequency-compositions joint PDF model provides a complete closure forturbulent reacting flows. Mass-weighted joint PDF is defined as the probability density functionof the simultaneous events of velocity, turbulent frequency and compositions at one time and onepoint. In the PDF methods, the transport equation for mass-weighted joint PDF is directly derivedfrom the Navier-Stokes equations (Pope, 1985) and the unclosed terms are modeled throughconstruction of stochastic differential equations (SDEs). The closure is usually guided by existentReynolds stress models such that the joint PDF model is equivalent to the correspondingReynolds stress model at the second moment level. Although advanced velocity models exist, thesimplified Langevin model (SLM) has been usually used in the PDF computations due to itssimplicity. In the present study, in addition to the SLM, we also consider two other popularvelocity models, namely Lagrangian Isotropization of Production Model (LIPM) and LagrangianSpeziale Sarkar Gatski Model (LSSG) and evaluated their performances for the bluff-body flows.Due to simplicity a flamelet approach will be used in the computations. Although mean fields inreacting bluff-body flow are found less sensitive to model constants it has been found that meanfields are very sensitive in non-reacting bluff-body flow. With the modification of LIPM andLSSG a significant improvement is achieved in prediction of mean fields for non-reacting bluff-body flow.Advisor: Metin Muradoğlu Date:Director: Prof. Dr. Yaman Arkun Date:
Sanal ortamlarda haptik geri beslemeli katı moleküler kenetlenme
In this thesis, we present computationally efficient methods for visualization andsimulation of molecular interactions in virtual environments with haptic feedback. In oursimulations, the haptic device is used to guide a rigid ligand molecule into a receptor sitewhile the molecular forces acting on the ligand molecule are scaled and reflected to the userin real-time. We demonstrate that the presence of a haptic interface accelerates the bindingprocess and reduce the binding errors if it is used as a precursor to estimate the initialconfiguration of the ligand molecule at the binding site. After placement and rough alignmentof the ligand molecule inside the binding cavity with the help of a haptic device, we use anovel computational approach to determine the final binding configuration of the ligandmolecule. In this approach, the ligand molecule is treated as a rigid body seeking for thelowest potential energy configuration. The rigid body configurations of the ligand moleculeis calculated in a least square sense using the new positions of its atoms moving under theinfluence of molecular interaction forces. The proposed approach is computationally moreefficient than the molecular simulation methods that utilize the standard rigid-bodyformulations. We also present new methods for haptic visualization of a protein surfaceinteractively to search for potential binding sites. Our experimental studies with 6 subjectsshow that subjects can successfully identify the true binding site among the 5 potentialbinding sites using visual and haptic cues. In addition, we show that the proposed distanceminimization approach can be used to find the final configuration of a ligand molecule insidethe binding cavity after it is initially aligned by the subject.
Kuantum hesaplama için bloch vektör formalismi
The Bloch vector formalism provides a useful approach to both fundamental andapplied problems of the finite dimensional, composite quantum systems, andtherefore it is a very important tool for the research on quantum computation andinformation. This thesis deals with the systematic treatment of Bloch vectorformalism for quantum mechanics of one node, two node and three node systems.Each chapter is divided into three sections dealing with state vectors, measurementsand time evolutions. Also, the geometric picture of the space of states will be givenin the section on one node systems. Classically correlated states i.e. separablestates, for two and three node systems will be investigated in the relevant chapters. Iwill also briefly mentioned the well-known EPR paradox, Bell's inequality, Mermin'sproofs on the problems with the elements of reality by using GHZ state and the twodifferent classes of the genuine tripartite entanglement to comprehend the unusualnature of entanglement.Advisor: Prof. Tekin Dereli Date:Director: Prof. Süleyman Özekici Date: