FBE- Bilgisayar Mühendisliği Lisansüstü Programı - Yüksek Lisans

Bu koleksiyon için kalıcı URI

Gözat

Son Başvurular

Şimdi gösteriliyor 1 - 5 / 328
  • Öge
    An enhanced two phase commit protocol for high performance consistency management in replicated state machines
    (Institute of Science and Technology, 2020-07) Uyanık, Halit ; Ovatman, Tolga ; 637248 ; Computer Engineering Department
    State Machines are used to represent a set of transitions, which makes it a useful tool for representing life-cycle of a specific set of events. By utilizing this property, a state machine can be replicated into several machines and each one of them can communicate with one another in order to keep track of the order of changes. This is called the replicated state machine approach and it is highly used in replicated data services where there is a need to manage the consistency of a system. In order to provide any consistency, it is necessary to use a communication algorithm which provides both high throughput, and less number of failures in the case of conflicting operations. One of the widely known communication protocols used in our context is the two-phase commit (2PC) protocol. It provides a two step algorithm in order to manage the committing actions between different machines for the same resource. First it checks if every machine in a network is ready for writing operation, then if a machine receives a successful message from all other machines, it will then proceed to commit the specific operation to all of them. Finally it applies the commit to its own resource. In the case of no priority between the writing actions between different machines, algorithm gives the commit rights to the first machine which can successfully receive OK from all others. However, when priority comes into action, and it becomes necessary to cancel out transitions with less importance, algorithm starts to cancel out some incoming transitions, and in most cases, if the writing operations are too frequent, it cancels out a writing operation even if it obtains its OK from other machines. Disadvantage of the common 2PC algorithm in the case of priority introduction, due to its phases follow one another without any transitions, when an incoming writing request fails, it has to repeat all the preceding events from that point again. When the distance between the last important point of no-return, such as reading a value into cache, and the point of 2PC protocol becomes further away, this affect is increased and the number of messages for a successful transition is increased. In order to reduce the overhead introduced by this problem, a new algorithm is implemented by enhancing the existing 2PC algorithm. Both steps of the 2PC algorithm mentioned is separated from one another, and can be freely deployed in any place on a state machine, as long as their order is preserved.
  • Öge
    Etmen tabanlı otomatik müzayede ortamının tasarım ve gerçeklenmesi
    (Fen Bilimleri Enstitüsü, 2020) Güleryüz, Uğur ; Erdoğan, Takuhi Nadia ; 636880 ; Bigisayar Mühendisliği
    Teknolojinin gelişmesi ile beraber internetin yaygınlaşmasından önce reel olarak yapılan pek çok işlem sanal ortama taşınmış ve daha az efor harcayarak yapılabilir hale gelmiştir. Günümüzde alışverişler büyük oranda online olarak yapılmaya başlanmıştır, satıcılar ürünlerini farklı yöntemler kullanarak satabilmektedir. Müzayedeler ürün satışı yapmak için kullanılan yaygın yöntemlerden birisidir ve günümüzde pek çok ürün farklı müzayede yöntemleri ile internet üzerinden satılabilmektedir. Farklı müzayede yöntemleri ile ürün satışı yapılırken hem alıcı hem de satıcı tarafının müzayedenin her aşamasını dikkatle takip etmesi ve ürünü satın alma isteği ve bütçesine göre doğru stratejilerle hareket etmesi son derece önemlidir. Fakat müzayedeyi takip etmek için yeterli zamanı olmayan veya aynı anda birden fazla müzayedeye alıcı veya satıcı olarak katılmış katılımcılar için müzayedeyi takip etme işlemi oldukça zor hale gelmektedir. Bu sebeple kullanıcının önceden belirlediği parametrelerle hem alıcı hem de satıcı tarafında müzayede takibini kullanıcı yerine otomatik yapacak bir sisteme ihtiyaç duyulmaktadır. Bu çalışmada kullanıcıların farklı tip müzayede yöntemleri kullanarak ürün alma ve ürün satma işlemlerini isterse manual isterse önceden gireceği parametre ve stratejilere göre otomatik olarak yapabileceği bir ortam tasarlanmış ve gerçeklenmiştir. Bu ortam tamamen etmen tabanlı olarak gerçeklenmiş ve uygulamanın ana bileşenleri etmenlerle ifade edilmiştir. Server, arayüz, teklif, müzayede olmak üzere dört tip etmen bulunmakta ve her etmen kendi temsil ettiği bileşenle alakalı işleri gerçekleştirmektedir. Farklı etmen tipleri kendi aralarında haberleşerek sistemin ana akışını oluşturmaktadır. Sistem İngiliz ve Hollandalı tipi müzayedeleri desteklemektedir ve her müzayede tipi için alıcı ve satıcı tarafına farklı stratejiler sunmaktadır. Hollandalı tipi müzayedede ücret değişimlerini yapan satıcı tarafına farklı ücret düşürme stratejileri sunmaktadır. Alıcı tarafına ise otomatik olarak müzayedeyi takip ederek istediği ücrete düştüğünde ürünü satın alma seçeneği sunmaktadır. İngiliz tipi müzayedede ise satıcı tarafına istediği ücrete ulaştığında otomatik olarak ürünü satma seçeneği sunmadktadır. Alıcı tarafına ise ister otomatik olarak belirlediği stratejide periyodik olarak teklifler verme ister belirlediği zamanda belirlediği miktarda teklif yapma seçeneği sunmaktadır. Yapılan deneylerde sistemin sunduğu seçeneklerle iki tip müzayede tipi içinde kullanıcının isteklerini karşıladığı, alıcıların ve satıcıların bütçeleri, ürünü alma veya satma isteklerine göre belirlediği stratejiler ve girdiği parametrelerle müzayedelere katılım sağlayabildiği gözlemlenmiştir. Sunulan stratejiler karşılaştırılmış olup farklı durumlarda farklı stratejilerin öne çıktığı ve müzayedenin sonucuna büyük oranda etki ettiği gözlemlenmiştir.
  • Öge
    Deshufflegan: Self-supervised learning for generative adversarial networks
    (Institute of Science and Technology, 2020-07) Baykal Can, Gülçin ; Ünal, Gözde ; 637455 ; Computer Engineering Programme
    Generative Adversarial Networks (GANs) attracted the attention of the research community with its performance in high quality image generations. After the idea of two player game theory as well as the multi-objective and multi-task loss ideas are introduced with the GAN models, numerous modifications on the architectures of the generator and the discriminator networks and the learning objectives are proposed. The basic intuition behind the desired improvements is to increase the quality of the generations at the output of the generator network of the GAN model. One of the ways to improve the generation performance is to enhance the discriminator network of the GAN model in order to learn expressive features of the real data and feed that information back to the generator of the GAN model. Original conditional GANs support the discriminator by adding the information of the class label as input along with the data. Class label information can be helpful as an additional signal to the training or the information can be used as a new task for the discriminator in order to increase its representation capacity. The capacity of the discriminator needs to be enhanced in order to learn meaningful features that can be used to distinguish between the real data and the fake data. As the usage of class labels improves the discriminator performance, equivalently the generation performance by the generator, this information can be beneficial in the training of GANs. However, as the acquirement of class labels is expensive in terms of both time and human resources, new ways of creating and incorporating additional information about the data should be considered. Self-supervised learning is a method to make use of the pseudo-labels of the data where these labels are obtained through an automatic process which is computationally light and easy. For example, the image can be rotated by 4 different degrees and the rotation degree can be used as a label for the data. Other than this, the input can be divided into pieces and the pieces can be shuffled. Then, the shuffling order can be treated as an additional information about the data. In this work, we propose a new method called DeshuffleGAN that deploys the additional task of deshuffling a shuffled image to the discriminator network of the GAN in order to enrich the learnt features by the discriminator. In order to perform deshuffling, structural relations among image tiles should be learnt. This implies that the discriminator should learn structurally coherent features of the data. As the generator tries to trick the discriminator by the synthesized images so that the discriminator treats them as the real data, the image generation quality should be improved such that the discriminator cannot distinguish them even with the learnt structural features. Therefore, the deshuffling task also supports the generator network to synthesize structurally coherent images. DeshuffleGAN outperforms the baseline methods demonstrated in this thesis and achieves both numerically and visually better results. We use FID calculation as the numerical evaluation metric where lower FID values imply the generated data distribution is similar to the real data distribution which is the desired outcome. We show that the DeshuffleGAN achieves lower FID values on datasets such as LSUN-Bedroom and LSUN-Church. We also use CelebA-HQ and CAT datasets and observe that self-supervision tasks may not always show significant effects on the generation quality of GANs. We further show the effects of the deshuffling task by employing different GAN architectures, and discuss which kind of discriminator architecture may be more appropriate to be coupled with a self-supervision task.
  • Öge
    Self-supervised pansharpening: Guided colorization of panchromatic images using generative adversarial networks
    (Institute of Science and Technology, 2020-07) Özçelik, Furkan ; Ünal, Gözde ; 637233 ; Bilgisayar Mühendisliği Bilim Dalı
    Satellite images provide images with different properties. Multispectral images have low spatial resolution and high spectral resolution. Panchromatic images have high spatial resolution and low spectral resolution. The fusion process of these two images is called pansharpening. For decades, traditional image processing methods are designed for this process. After the inspirational success of Convolutional Neural Networks(CNN) in computer vision, CNN models are also designed for pansharpening. Convolutional Neural Networks (CNN)-based approaches have shown promising results in pansharpening of satellite images in recent years. However, they still exhibit limitations in producing high-quality pansharpening outputs. We identified a spatial detail disagreement problem between reduced resolution panchromatic images and original multispectral images, which are assumed to have the same resolution. This problem causes an insufficient training process in current CNN-based pansharpening models. We propose a new self-supervised learning framework, where we treat pansharpening as a colorization problem, which brings an entirely novel perspective and solution to the problem compared to existing methods that base their solution solely on producing a super-resolution version of the multispectral image. CNN-based methods provide a reduced resolution panchromatic image as input to their model along with reduced resolution multispectral images, hence learn to increase their resolution together. In the training phase of our model, reduced resolution panchromatic image is substituted with grayscale transformed multispectral image, thus our model learns colorization of the grayscale input. We further address the fixed downscale ratio assumption during training, which does not generalize well to the full-resolution scenario. We introduce a noise injection into the training by randomly varying the downsampling ratios. Those two critical changes, along with the addition of adversarial training in the proposed PanColorization Generative Adversarial Networks (PanColorGAN) framework, help overcome the spatial detail loss and blur problems that are observed in CNN-based pansharpening. The proposed approach outperforms the previous CNN-based and traditional methods as demonstrated in our experiments.
  • Öge
    Deep hybrid recommender system
    (Institute of Science And Technology, 2020-03-11) Türker, Didem ; Öğüdücü, Şule ; 504161537 ; Computer Engineering ; Bilgisayar Mühendisliği
    With the increasing popularity of e-commerce platforms in recent years, recommendation systems have become highly popular. E-commerce platforms can offer items to the user with personalized information from large quantities of data that are often dirty and difficult to use. Not only e-commerce platforms, but also many social media and many platforms where users interact with items use recommendation systems. Recommendation systems also provide a quality user experience to users. In this way, users can easily reach the items according to their taste instead of getting lost among many products. Traditional methods mainly based on the user-item interactions for recommendation. However user-item interactions mostly suffers from data sparsity problem. Data sparsity is the term used to describe the fact that not observing enough data. For example, recommender systems recommend thousands of products to hundreds of thousands of users, if you stored the data about user-product interaction in a matrix, it would be a huge amount of data consisting of lots of zeros. In addition, when a new user with no interaction with any item or a new item with no interaction with any user is included in the system, recommendation cannot be generated for that user or item using only user-item interactions. Therefore, beyond the user-item interactions, rich side information is a good source to increase the quality of recommendation. To mitigate the sparsity issue and improve the recommendation quality, we incorporate side information with user-item interactions. In recommender systems, explicit or implicit feedback of users is used. Implicit feedback (purchase/nonpurchase or click/nonclick etc.) represents opinion indirectly through analyzing user behaviour. Using implicit feedback is more challenging because of lack of negative feedback. But the high-quality explicit feedback, which provides direct input from users about their item interests, is the most functional input type to understand the user's exact response for an item. However, many systems do not have explicit feedback and it is difficult to collect this type of feedback. We have tested our framework with real world fashion retailer e-commerce data using implicit feedback. In our study, item purchasing is considered as positive feedback, and negative feedback is randomly selected from unobserved interactions. Consequently, lack of negative feedback of the implicit feedback was tried to be eliminated. For long years, neural networks have significant role in many areas of computer science and have gained popularity in recommendation systems in recent years. Successful results have been obtained in understanding the complex and non-linear relationship between user-item interactions with neural networks. In this study, artificial neural networks are used to increase the performance in understanding the complex and nonlinear relationship between user-item interactions. Also, user-item interactions are combined with side information to solve the problem of data sparsity and improve the recommendation performance. In this study, two neural network architecture are proposed. In the first proposed model, our input is purchasing count of product. Mean Squared Error (MSE) is optimized and recommendation task is performed as rating prediction. Also MSE is used as evaluation metric of the first model. In the second proposed model, the output of the model is binary classification since input of this model is taken as 1 for positive feedback or 0 for negative feedback, then binary cross entropy loss is optimized. Top-k recommendation is made instead of rating prediction. Hit Ratio (HR) is used to evaluate second model.