Bilgisayar Mühendisliği Bölümü Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.11779/1940

Browse

Now showing 1 - 20 of 110

Citation - Scopus: 2
Turcoins: Turkish Republic Coin Dataset
(IEEE, 2021) Gökberk, Berk; Akarun, Lale; Temiz, Hüseyin
In this paper, we present a novel and comprehensive dataset which contains Turkish Republic coins minted since 1924 and present a deep learning based system that can automatically classify coins. The proposed dataset consists of 11080 coin images from 138 different classes. To classify coins, we utilize a pre-trained neural network (ResNet50) which is pre-trained on ImageNet. We train the pre-trained neural networks on our dataset by transfer learning. The imbalanced nature of the dataset causes the classifier to show lower performance in classes with fewer samples. To alleviate the imbalance problem, we propose a StyleGAN2-based augmentation method providing realisticfake coins for rare classes. The dataset will be published in http://turcoins.
Citation - WoS: 3
Citation - Scopus: 4
A Novel Graph Transformation Strategy for Optimizing Sptrsv on Cpus
(Wiley, 2023) Yılmaz, Buse
Sparse triangular solve (SpTRSV) is an extensively studied computational kernel. An important obstacle in parallel SpTRSV implementations is that in some parts of a sparse matrix the computation is serial. By transforming the dependency graph, it is possible to increase the parallelism of the parts that lack it. In this work, we present a novel graph transformation strategy to increase the parallelism degree of a sparse matrix and compare it to our previous strategy. It is seen that our transformation strategy can provide a speedup as high as 1.42x$$ 1.42x $$.
Optimizing Collective Building Management Through a Machine Learning-Based Decision Support System
(IEEE, 2023) Güvençli, Mert; Dağ, Hasan; Doğan, Erkan; Çakar, Tuna; Özyürüyen, Burcu; Kiran, Halil
This study presents the design, implementation, and evaluation of a Decision Support System (DSS) developed for Collective Building Management. Given the potential advantages of machine learning techniques in this domain, the research explores how these techniques can be used to improve collective building management. The dataset consists of 824,932 records and 15 attributes, after preprocessing the data to fill in missing values with the median. The random forest algorithm was chosen for model training and achieved a performance rate of 71.2%. This model can be used to optimize decision processes in collective building management. The proposed prototype is notable for its ability to automatically generate operational plans. In conclusion, machine learning-based DSSs are effective tools for collective building management.
Özyinelemeli Sinir Ağları ile Türkçe Doğal Dil Üretimi
(TÜBİTAK, 2018) Demir, Şeniz; Gökmen, Muhittin
İnsanlar arasındaki iletişimi sağlayan doğal diller, zaman içinde insanlarla etkin ve kullanıcı dostu etkileşim kurabilmek amacıyla sistemler ve yazılımlar tarafından kullanılmaya başlanmıştır. Tıpkı insanlar gibi sesli veya yazılı doğal dil ifadelerini anlayabilen ve sonrasında kullanıcıların beklentilerini karşılayabilen dil tabanlı teknolojiler (örn. arama motorları, bilgisayar destekli eğitici sistemler ve diyalog sistemleri) bu motivasyonla ortaya çıkmıştır. Bu çalışmalarda, problemin doğası ve hedef dilin yapısındaki zorluklara ek olarak insanların doğal dilleri nasıl öğrendiğini ve kullandığını modellemedeki kısıtlar başarım oranlarını etkilemiştir. Günümüzde, dil tabanlı teknolojiler insanlar tarafından yaygın şekilde kullanılıyor olsalar da (örn. Google Arama Motoru ve Apple Siri), ulaşılan teknolojik seviye hedef dile göre çeşitlilik göstermektedir. Sondan eklemeli ve zengin dil yapısı ile Türkçe geliştirilen teknolojik çözümler ve üretilen veri kaynakları açısından pek çok doğal dilin gerisinde kalmaktadır. Ayrıca, bugüne kadar Türkçe dil teknolojileri konusunda yapılan çalışmaların ağırlıklı olarak dili işleme, anlama ve analiz etmeye dönük (örn. kelimelerin morfolojik analizi, özel isim tespiti, bağlılık çözümlemesi, metin sınıflandırma ve metin özetleme) olduğu gözlemlenmektedir. Türkçe dil üretimi konusunda sınırlı yeteneklere sahip ve akademik seviyede kalarak devamı getirilmemiş birkaç çalışma mevcuttur. Fakat bu çalışmalar karmaşık sayılabilecek dilbilimi teorileri ile ifade edilen içerik ifadelerini cümlelere dönüştürmekten öteye geçmemiştir ve başka uygulamalarla entegre olarak test edilmemiştir. Bu çalışmada, Türkçe dilinin derin öğrenme tabanlı bir sistem (dil aracı) ile otomatik olarak üretimi hedeflenmektedir. Bu sistemin, girdi olarak verilen içerik ifadelerini Türkçe dili kurallarına uygun ve anlaşılır cümlelere dönüştüreceği öngörülmektedir. Literatürdeki en kapsamlı Türkçe dil üretimi sistemi olması planlanan bu çalışmada son yıllarda pek çok dil teknolojisinde başarımı ispat edilmiş diziden diziye öğrenebilen (örn. kelime dizisinden başka bir kelime dizisi) özyinelemeli sinir ağı yapıları kullanılacaktır. Bu ağların sağladığı dinamiklik ile farklı çeşitler (örn. uzun kısa süreli bellek ve girişli özyinelemeli birim) ve genişlemeler (örn. dikkat mekanizması) denenecektir ve başarımı en yüksek sinir ağı mimarisi belirlenecektir. Buna ek olarak, sinir ağlarının kullanımı bazı faktörlerin (örn. bağlam bilgisi ve kullanıcı tercihleri) sisteme entegrasyonuna ve üretim aşamasına olan etkilerinin incelenmesine imkân sağlayacaktır.
Citation - WoS: 3
Citation - Scopus: 2
A Benchmark Dataset for Turkish Data-To Generation
(Elsevier, 2022) Demir, Şeniz; Öktem, Seza
In the last decades, data-to-text (D2T) systems that directly learn from data have gained a lot of attention in natural language generation. These systems need data with high quality and large volume, but unfortunately some natural languages suffer from the lack of readily available generation datasets. This article describes our efforts to create a new Turkish dataset (Tr-D2T) that consists of meaning representation and reference sentence pairs without fine-grained word alignments. We utilize Turkish web resources and existing datasets in other languages for producing meaning representations and collect reference sentences by crowdsourcing native speakers. We particularly focus on the generation of single-sentence biographies and dining venue descriptions. In order to motivate future Turkish D2T studies, we present detailed benchmarking results of different sequence-to-sequence neural models trained on this dataset. To the best of our knowledge, this work is the first of its kind that provides preliminary findings and lessons learned from the creation of a new Turkish D2T dataset. Moreover, our work is the first extensive study that presents generation performances of transformer and recurrent neural network models from meaning representations in this morphologically-rich language.
Comparing Humans and Deep Neural Networks on Face Recognition Under Various Distance and Rotation Viewing Conditions
(Journal of Vision, 2023) Fux, Michal; Arslan , Şuayb Şefik; Jang, Hojin; Boix, Xavier; Cooper, Avi; Groth, Matt J; Sinha, Pawan
Humans possess impressive skills for recognizing faces even when the viewing conditions are challenging, such as long ranges, non-frontal regard, variable lighting, and atmospheric turbulence. We sought to characterize the effects of such viewing conditions on the face recognition performance of humans, and compared the results to those of DNNs. In an online verification task study, we used a 100 identity face database, with images captured at five different distances (2m, 5m, 300m, 650m and 1000m) three pitch values (00 - straight ahead, +/- 30 degrees) and three levels of yaw (00, 45, and 90 degrees). Participants were presented with 175 trials (5 distances x 7 yaw and pitch combinations, with 5 repetitions). Each trial included a query image, from a certain combination of range x yaw x pitch, and five options, all frontal short range (2m) faces. One was of the same identity as the query, and the rest were the most similar identities, chosen according to a DNN-derived similarity matrix. Participants ranked the top three most similar target images to the query image. The collected data reveal the functional relationship between human performance and multiple viewing parameters. Nine state-of-the-art pre-trained DNNs were tested for their face recognition performance on precisely the same stimulus set. Strikingly, DNN performance was significantly diminished by variations in ranges and rotated viewpoints. Even the best-performing network reported below 65% accuracy at the closest distance with a profile view of faces, with results dropping to near chance for longer ranges. The confusion matrices of DNNs were generally consistent across the networks, indicating systematic errors induced by viewing parameters. Taken together, these data not only help characterize human performance as a function of key ecologically important viewing parameters, but also enable a direct comparison of humans and DNNs in this parameter regime
Cooperative Network Coding for Distributed Storage Using Base Stations With Link Constraints
(arXiv, 2021) Arslan, Şuayb Şefik; Pourmandi, Massoud; Haytaoğlu, Elif
In this work, we consider a novel distributed data storage/caching scenario in a cellular setting where multiple nodes may fail/depart at the same time. In order to maintain the target reliability, we allow cooperative regeneration of lost nodes with the help of base stations allocated in a set of hierarchical layers. Due to this layered structure, a symbol download from each base station has a different cost, while the link capacities connecting the nodes of the cellular system and the base stations are also limited. In this more practical and general scenario, we present the fundamental trade-off between repair bandwidth cost and the storage space per node. Particularly interesting operating points are the minimum storage as well as bandwidth cost points in this trade-off curve. We provide closed-form expressions for the corresponding bandwidth (cost) and storage space per node for these operating points. Finally, we provide an explicit optimal code construction for the minimum storage regeneration point for a given set of system parameters.
Citation - WoS: 1
Citation - Scopus: 1
Fuzzy Elephant Herding Optimization and DBSCAN for Emergency Transportation: A Case Study for the 2023Turkiye Earthquake
(Springer international Publishing Ag, 2024) Drias, Yassine; Drias, Habiba
In recent times, our planet has experienced numerous natural disasters across all continents. The damage caused by these disasters has been so extensive that Emergency Medical Services (EMS) proved incapable of handling the situation. In this article, we present a novel approach for urgent disaster transport with the aim of minimizing loss of life. In this context, we are investigating the Density-Based Spatial Clustering of Applications with Noise algorithm (DBSCAN) to cluster the large geographic zone affected by the 2023 earthquake in Turkiye. The clustering is done based on hospitals' capacity on one hand and damages on the other hand. The ambulance dispatching task is then tackled using a new fuzzy version of Elephant Herding Optimization called FEHO. This approach addresses the challenge of dispatching ambulances to cover emergency locations effectively and optimally in the clustered regions. Experiments conducted on real data demonstrate the effectiveness of our approach in managing emergency transportation and highlight its potential to minimize the number of casualties.
Citation - WoS: 1
Citation - Scopus: 1
A Novel Genetic Algorithm-Based Improvement Model for Online Communities and Trust Networks
(IOS Press, 2020) Bekmezci, ilker; Cimen, Egemen Berkic; Ermiş, Murat
Social network analysis offers an understanding of our modern world, and it affords the ability to represent, analyze and even simulate complex structures. While an unweighted model can be used for online communities, trust or friendship networks should be analyzed with weighted models. To analyze social networks, it is essential to produce realistic social models. However, there are serious differences between social network models and real-life data in terms of their fundamental statistical parameters. In this paper, a genetic algorithm (GA)-based social network improvement method is proposed to produce social networks more similar to real-life data sets. First, it creates a social model based on existing studies in the literature, and then it improves the model with the proposed GA-based approach based on the similarity of the average degree, the k-nearest neighbor, the clustering coefficient, degree distribution and link overlap. This study can be used to model the structural and statistical properties of large-scale societies more realistically. The performance results show that our approach can reduce the dissimilarity between the created social networks and the real-life data sets in terms of their primary statistical properties. It has been shown that the proposed GA-based approach can be used effectively not only in unweighted networks but also in weighted networks.
Power Savings in Cold Storage
(2016) Arslan, Şuayb Şefik; Göker, Turguy
Methods and apparatus associated with data cold storage are described. Example apparatus include an array of data storage devices arranged in rows and columns. Columns of the array are orthogonal to rows. A row has an associated row-centric power supply, and a column has an associated column-centric local electronics module (LEM) that controls a data storage device in the column independently of other data storage devices in the array. Example apparatus include logics that control a power mode of a data storage device independently of other data storage devices in the array, that control a power mode of an LEM, that adaptively regulate the level of data stored in a buffer, and that determine whether a data object will be stored in the buffer or stored on a data storage device in the array, based on the probability the data object will be accessed within a threshold period of time.
Citation - WoS: 2
Citation - Scopus: 5
Mojette Transform Based Ldpc Erasure Correction Codes for Distributed Storage Systems
(2017) Arslan, Şuayb Şefik; Normand, Nicolas; Parrein, Benoit
Mojette Transform (MT) based erasure correction coding possesses extremely efficient encoding/decoding algorithms and demonstrate promising burst erasure recovery performance. MT codes are based on discrete geometry and provide redundancy through creating projections. Projections are made of smaller data structures called bins and are generated from a two dimensional convex-shaped data. For exact data recovery, only a subset of projections are needed by the decoder. We realize that the discrete geometry definition of MT erasure codes corresponds to creating structured/deterministic generator matrices. In this study, we show an alternative Low Density Parity Check (LDPC) code construction methodology through investigating parity check matrices of MT codes which shows sparseness as the blocklength of the code gets large. In a distributed storage setting, we also quantify the repair bandwidth and show that this novel interpretation can be used to facilitate bin-level local repairs.
Citation - WoS: 2
Citation - Scopus: 2
Distributed Matrix Multiplication With Mds Array Bp-Xor Codes for Scaling Clusters
(IEEE, 2019) Arslan, Şuayb Şefik
This study presents a novel coded computation technique for distributed matrix-matrix product computation at a massive scale that outperforms well known previous strategies in terms of total execution time. Our method achieves this performance by distributing the encoding operation over the cluster (slave) nodes at the expense of increased master-slave communication. The product computation is performed using MDS array Belief Propagation (BP)-decodable codes based on pure XOR operations. In addition, our scheme is configurable and suited for modern compute node architectures equipped with multiple processing units organized in a hierarchical manner. Assuming the number of backup nodes being sublinear in the size of the product, we shall demonstrate that the proposed scheme achieves order-optimal computation from an end-to-end latency perspective while ensuring acceptable communication requirements that can be addressed by today's high speed network link infrastructures.
Citation - WoS: 6
Citation - Scopus: 6
Efficient Strategy for Multi-Uav Path Planning in Target Coverage Problems
(IEEE, 2022) Bekmezci, İlker; Pehlivanoğlu, Perihan; Pehlivanoğlu, Y. Volkan
In recent years, multi unmanned aerial vehicles (UAVs) are used in the same system to accomplish more complex missions. In many multi-UAV system applications, the main objective is to visit some predetermined checkpoints in operational area. If the number of check points and constraints increases, finding a feasible solution takes up too much time. In this paper, a checkpoint based multi-UAV path planning problem is solved by using improved genetic algorithm. The main contributions of this paper are: (1) the introducing revisit time interval concept, (2) the investigating of the effect of objective function description, and (3) looking into an outcome of using multiple runways on optimal multi-UAV path planning. The proposed strategy-based optimization methodology is performed for checkpoint based multi-UAV path planning problems in two-dimensional (2D) environment. Performance results show that the proposed strategy provides effective and feasible paths for each UAV.
A New Benchmark Dataset for P300 Erp-Based Bci Applications
(Academic Press Inc Elsevier Science, 2023) Çakar, Tuna; Özkan, Hüseyin; Musellim, Serkan; Arslan, Suayb S.; Yağan, Mehmet; Alp, Nihan
Because of its non-invasive nature, one of the most commonly used event-related potentials in brain -computer interface (BCI) system designs is the P300 electroencephalogram (EEG) signal. The fact that the P300 response can easily be stimulated and measured is particularly important for participants with severe motor disabilities. In order to train and test P300-based BCI speller systems in more realistic high-speed settings, there is a pressing need for a large and challenging benchmark dataset. Various datasets already exist in the literature but most of them are not publicly available, and they either have a limited number of participants or utilize relatively long stimulus duration (SD) and inter-stimulus intervals (ISI). They are also typically based on a 36 target (6 x 6) character matrix. The use of long ISI, in particular, not only reduces the speed and the information transfer rates (ITRs) but also oversimplifies the P300 detection. This leaves a limited challenge to state-of-the-art machine learning and signal processing algorithms. In fact, near-perfect P300 classification accuracies are reported with the existing datasets. Therefore, one certainly needs a large-scale dataset with challenging settings to fully exploit the recent advancements in algorithm design (machine learning and signal processing) and achieve high-performance speller results. To this end, in this article we introduce a new freely-and publicly-accessible P300 dataset obtained using 32-channel EEG, in the hope that it will lead to new research findings and eventually more efficient BCI designs. The introduced dataset comprises 18 participants performing a 40 -target (5 x 8) cued-spelling task, with reduced SD (66.6 ms) and ISI (33.3 ms) for fast spelling. We have also processed, analyzed, and character-classified the introduced dataset and we presented the accuracy and ITR results as a benchmark. The introduced dataset and the codes of our experiments are publicly accessible at https://data .mendeley.com /datasets /vyczny2r4w.(c) 2023 Elsevier Inc. All rights reserved.
Erasure Coding Magnetic Tapes for Minimum Latency and Adaptive Parity Protection Feedback
(Patent Ofisi : US, 2019) Goker, Turguy; Arslan, Şuayb Şefik; Le, Hoa; Peng, James; Prigge, Carsten
A magnetic tape device or system can store erasure encoded data that generates a multi-dimensional erasure code corresponding to an erasure encoded object comprising a code-word (CW). The multi-dimensional erasure code enables using a single magnetic tape in response to a random object/file request, and correct for an error within the single magnetic tape without using other tapes. Encoding logic can further utilize other magnetic tapes to generate additional parity tapes that recover data from an error of the single magnetic tape in response to the error satisfying a threshold severity for a reconstruction of the erasure coded object or chunk (s) of the CW. The encoding logic can be controlled, at least in part, by one or more iterative coding processes between multiple erasure code dimensions that are orthogonal to one another.
Citation - WoS: 27
Citation - Scopus: 30
Service-Aware Multi-Resource Allocation in Software-Defined Next Generation Cellular Networks
(2018) Arslan, Şuayb Şefik; Zeydan, Engin; Narmanloğlu, Ömer
Network slicing is one of the major solutions needed to meet the requirements of next generation cellular networks, under one common network infrastructure, in supporting multiple vertical services provided by mobile network operators. Network slicing makes one shared physical network infrastructure appear as multiple logically isolated virtual networks dedicated to different service types where each Network Slice (NS) benefits from on-demand allocated resources. Typically, the available resources distributed among NSs are correlated and one needs to allocate them judiciously in order to guarantee the service, MNO, and overall system qualities. In this paper, we consider a joint resource allocation strategy that weights the significance of the resources per a given NS by leveraging the correlation structure of different quality-of-service (QoS) requirements of the services. After defining the joint resource allocation problem including the correlation structure, we propose three novel scheduling mechanisms that allocate available network resources to the generated NSs based on different type of services with different QoS requirements. Performance of the proposed schedulers are then investigated through Monte-Carlo simulations and compared with each other as well as against a traditional max-min fairness algorithm benchmark. The results reveal that our schedulers, which have different complexities, outperform the benchmark traditional method in terms of service-based and overall satisfaction ratios, while achieving different fairness index levels.
Citation - WoS: 20
Citation - Scopus: 28
An Evaluation of Recent Neural Sequence Tagging Models in Turkish Named Entity Recognition
(Elsevier, 2021) Makaroğlu, Didem; Demir, Şeniz; Aras, Gizem; Çakır, Altan
Named entity recognition (NER) is an extensively studied task that extracts and classifies named entities in a text. NER is crucial not only in downstream language processing applications such as relation extraction and question answering but also in large scale big data operations such as real-time analysis of online digital media content. Recent research efforts on Turkish, a less studied language with morphologically rich nature, have demonstrated the effectiveness of neural architectures on well-formed texts and yielded state-of-the art results by formulating the task as a sequence tagging problem. In this work, we empirically investigate the use of recent neural architectures (Bidirectional long short-term memory (BiLSTM) and Transformer-based networks) proposed for Turkish NER tagging in the same setting. Our results demonstrate that transformer-based networks which can model long-range context overcome the limitations of BiLSTM networks where different input features at the character, subword, and word levels are utilized. We also propose a transformer-based network with a conditional random field (CRF) layer that leads to the state-of-the-art result (95.95% f-measure) on a common dataset. Our study contributes to the literature that quantifies the impact of transfer learning on processing morphologically rich languages.
Citation - WoS: 3
Citation - Scopus: 5
Unraveling Neural Pathways of Political Engagement: Bridging Neuromarketing and Political Science for Understanding Voter Behavior and Political Leader Perception
(2023) Çakar, Tuna; Filiz, Gözde
Political neuromarketing is an interdisciplinary field that combines marketing, neuroscience, and psychology to understand voter behavior and political leader perception. This interdisciplinary field offers novel techniques to understand complex phenomena such as voter engagement, political leadership, and party branding. This study aims to understand the neural activation patterns of voters when they are exposed to political leaders using functional near-infrared spectroscopy (fNIRS) and machine learning methods. We recruited participants and recorded their brain activity using fNIRS when they were exposed to images of different political leaders. This neuroimaging method (fNIRS) reveals brain regions central to brand perception, including the dorsolateral prefrontal cortex (dlPFC), the dorsomedial prefrontal cortex (dmPFC), and the ventromedial prefrontal cortex (vmPFC). Machine learning methods were used to predict the participants' perceptions of leaders based on their brain activity. The study has identified the brain regions that are involved in processing political stimuli and making judgments about political leaders. Within this study, the best-performing machine learning model, LightGBM, achieved a highest accuracy score of 0.78, underscoring its efficacy in predicting voters' perceptions of political leaders based on the brain activity of the former. The findings from this study provide new insights into the neural basis of political decision-making and the development of effective political marketing campaigns while bridging neuromarketing, political science and machine learning, in turn enabling predictive insights into voter preferences and behavior
Citation - Scopus: 7
The Use of Neurometric and Biometric Research Methods in Understanding the User Experience During Product Search of First-Time Buyers in E-Commerce - Conference Paper
(Springer, 2017) Rızvanoğlu, Kerem; Gürvardar, İrfan; Çakar, Tuna; Öztürk, Özgürol; Zengin Çelik, Deniz
Understanding user experience (UX) during e-commerce has been a relatively important research area especially in the last decade. The use of conventional methods in UX such as task-observation, in-depth interviews and questionnaires has already contributed for the measurement of the efficiency and effectiveness. This empirical study has aimed to make use of both conventional and neuroscientific methods simultaneously to provide a richer analysis framework for understanding the product search experience of the first-time buyers. The current work provides insights for the results from the combined use of conventional and neuroscientific-biometric methods in a UX study. Although this has been an exploratory study within a limited literature, the obtained results indicate a potential use of these methods for UX research, which may contribute to improve the relevant experience in various digital platforms.
Next-Generation Data Storage: Transistor and Dna
(Institute of Electrical and Electronics Engineers Inc., 2018) Pusane, Ali E.; Arslan, Şuayb Şefik; Ashrafi, Reza A.
With the generation of diverse data growing at exponential rates, investigating better digital storage media is inevitable. Currently, one solution is the utilization of solid-state based memory devices, which offer several desirable characteristics, including very fast write/read operations, scalability, and reduced fabrication costs. However, with the increased need for long term and large storage space, their data retention capabilities drastically decline. Another emerging storage technology on the horizon is the biotechnological based DNA storage, which renders a phenomenal storage capacities. In this paper, basics of these two promising storage technologies are reviewed and their potential future trends are discussed. © 2018 IEEE.

Browse

Browsing Bilgisayar Mühendisliği Bölümü Koleksiyonu by Language "en"