WoS İndeksli Yayınlar Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/394

Browse

Filters

2024 - 202523

Settings

Start date: 2024Publisher: search.filters.publisher.IEEE

Search Results

Now showing 1 - 10 of 23
  • Conference Object
    Citation - WoS: 1
    Citation - Scopus: 1
    Oscillator Phase Noise Impact on Monostatic/Bistatic Space-Borne Sub-THz ISAR
    (IEEE, 2025-05-21) Bekari, Ali; Gashinova, Marina; Bekar, Muge; Martorellai, Marco; Antonioni, Michail; Bekar, Ali; Martorella, Marco; Antoniou, Michail
    This study develops an oscillator phase noise model and analyzes its effects on the performance of spaceborne monostatic and bistatic Inverse Synthetic Aperture Radar (B-ISAR) systems operating at the sub-THz band. The B-ISAR study is of current importance as it can provide a basis for distributed space-based ISAR to enable persistent co-operative space domain awareness (Co-SDA).
  • Conference Object
    Enhancing Complex Disease Group Scoring with Mirgedinet: A Multi-Algorithm Machine Learning Framework Based on the GSM Approach
    (IEEE, 2025-06-25) Qumsiyeh, Emma; Bakir-Gungor, Burcu; Yousef, Malik
    Integrating biological prior knowledge for disease gene associations has shown significant promise in discovering new biomarkers with potential translational applications. This work investigates the application of a multi-algorithm machine learning framework based on the Grouping-Scoring-Modeling (G-S-M) approach for improving the prediction of complex diseases. The study identifies the primary gene and miRNA interactions in various complex diseases with the help of miRGediNET, which is a machine-learning based tool that integrates data from three biological databases. Traditional methods have only focused on independence between features; the G-S-M method focuses on aggregating genes based on biological interactions, pinpointing the scoring of gene groups for a disease, and modeling its predictive capability using advanced machine learning algorithms. In this research paper, seven algorithms, including Support Vector Machine, Decision Tree, and CatBoost, were applied to eight datasets extracted from the GEO database. This framework proved very robust in ranking gene clusters, thus predicting critical biomarkers while doing 100-fold randomized cross-validation within the evaluation. The results indicate this approach's high potential for refining disease and supporting research for choosing the best algorithm that can provide biological insights and computational advances.
  • Conference Object
    High Performance and Resource Efficient Low Density Parity Check Decoder Design
    (IEEE, 2025-06-25) Unal, Burak
    Low Density Parity Check (LDPC) codes have gained popularity in communication systems due to their capacity-approaching error correction performance. In this study, a highperformance LDPC decoding algorithm with extremely low resource usage is proposed. Among the hard decision class of LDPC decoders, Gallager B (GaB) provides high-performance hardware due to its computational simplicity. However, GaB suffers from poor error-correction performance. In this study, a new intrinsic computation technique for GaB called Intrinsic Gallager B (IGaB) is introduced to improve error correction performance. Our simulation results show that the IGaB algorithm provides better error correction performance compared with GaB. GaB and IGaB algorithms are implemented on Field Programmable Gate Array (FPGA) to compare hardware performance.
  • Conference Object
    Exploring Microbiome Signatures in Autism Spectrum Disorder via Grouping-Scoring Based Machine Learning
    (IEEE, 2025-06-25) Temiz, Mustafa; Ersoz, Nur Sebnem; Yousef, Malik; Bakir-Gungor, Burcu
    The rapid increase in omic data production increased the importance of machine learning (ML) methods to analze these data. In particular, the use of metagenomic data in the diagnosis, prognosis and treatment of diseases is becoming widespread. Autism Spectrum Disorder (ASD) is a neurodevelopmental disease that occurs in early childhood and continues lifelong. The aim of this study is to increase ML performance, reduce computational costs and achieve successful classification performance using a small number of metagenomic features. In addition, disease prediction is performed; ASD associated biomarkers are determined using the microBiomeGSM on metagenomic data. Classification is performed at three different taxonomic levels (genus, family and order) using the relative abundance values of species. The best performance metric (0.95 AUC) was obtained at the order taxonomic level using an average of 416 features with microBiomeGSM. The identified ASD-related taxonomic species are presented.
  • Conference Object
    Citation - WoS: 2
    Citation - Scopus: 2
    Fine Tuning DeepSeek and Llama Large Language Models with LoRA
    (IEEE, 2025-06-25) Uluirmak, Bugra Alperen; Kurban, Rifat
    In this paper, Low-Rank Adaptation (LoRA) finetuning of two different large language models (DeepSeek R1 Distill 8B and Llama3.1 8B) was performed using the Turkish dataset. Training was performed on Google Colab using A100 40 GB GPU, while the testing phase was carried out on Runpod using L4 24 GB GPU. The 64.6 thousand row dataset was transformed into question-answer pairs from the fields of agriculture, education, law and sustainability. In the testing phase, 40 test questions were asked for each model via Ollama web UI and the results were supported with graphs and detailed tables. It was observed that the performance of the existing language models improved with the fine-tuning method.
  • Conference Object
    Range-Based Wireless Sensor Network Localization by a Circumnavigating Mobile Anchor Without Position Information
    (IEEE, 2024-06-11) Guler, Samet
    Typical range-based wireless sensor network (WSN) localization approaches aim at estimating the sensor node positions by using a set of anchors with known positions. In some applications, assuming the knowledge of the anchors' positions may be impractical, and estimation of the sensors' positions in an arbitrary fixed frame may be sufficient. Considering such scenarios, we propose a WSN localization algorithm by single mobile anchor without self location information. The mobile anchor obtains distance measurements from the sensors while tracking a custom trajectory which is shown to improve the localization performance over time for high signal-to-noise ratio cases. By utilizing two stationary reference nodes within the WSN, the proposed framework generates sensor node position estimation up to translation and rotation with sufficient precision in the absence of global positioning aids. We foresee that the proposed framework can demonstrate benefits in several WSN applications ranging from internet-of-things to service robotics.
  • Conference Object
    Citation - WoS: 1
    Citation - Scopus: 1
    Prediction of Type 2 Diabetes Using Metagenomic Data and Identification of Taxonomic Biomarkers
    (IEEE, 2024-05-15) Temiz, Mustafa; Kuzudisli, Cihan; Yousef, Malik; Bakir-Gungor, Burcu
    Nowadays, different molecular levels of -omics data on diseases are generated and analyzing these data with machine learning methods is one of the popular research topics. Among these data, the use of metagenomic data to facilitate the diagnosis, detection and treatment of diseases is increasing day by day. Type 2 diabetes (T2D) is a chronic disease characterized by insulin resistance and progressive dysfunction of pancreatic beta cells. While the number of people with diabetes is increasing by around 8% annually, the cost of treating the disease is rising by 18% per year. Therefore, the number of studies on the diagnosis, development and progression of T2D is increasing over time. The aim of this study is to achieve higher machine learning performance by using fewer metagenomic features and to achieve better classification performance by reducing computational costs. In this study, we compare the performance of three different methods using T2D-related metagenomic data. First, the MetaPhlAn tool is used to calculate the taxonomic species and their relative abundances in each sample. The SVM-RCE, RCE-IFE and microBiomeGSM tools used in this study are methods that perform classification by grouping and scoring features and are known to work well on complex datasets. In this study, the best results were obtained with the RCE-IFE tool with an AUC of 0.72 with an average of 125 features information. In addition, key taxonomic species identified by these tools as associated with T2D are presented in comparison to the literature.
  • Conference Object
    Power Factor Improvement of a Permanent-Magnet Vernier Machine with Harmonic Injected Excitation Currents
    (IEEE, 2025-06-11) Karatepe, Hasan Can; Tekgun, Didem
    Permanent-magnet vernier machines (PMVM) are recognized for their high torque density but low power factor (PF) due to high inductive reactance. This paper presents a method for improving the PF of a PMVM by injecting additional harmonics into the excitation currents. This injection is done through the motor drive, unlike many proposed methods for enhancing PF, thus eliminating any modifications needed on the machine's geometric design. In this paper, different sets of harmonic injected currents are fed to a 14-rotor pole 12-slot PMVM with short-pitched coils on Finite Element Analysis (FEA) to demonstrate the effects of individual and combined harmonic currents. Corresponding performance characteristics of each simulation case, such as PF and torque density, are investigated. Simulation results indicate that PF can be improved by the proposed method of harmonic current injection. A comparison with a similarly sized permanent-magnet synchronous machine (PMSM) is made to demonstrate that the proposed method can be an alternative to widely used PMSMs.
  • Conference Object
    Citation - WoS: 2
    Citation - Scopus: 2
    Optimal Dead Band Control of Occupant Thermostats for Grid-Interactive Homes
    (IEEE, 2024-09-10) Savasci, Alper; Ceylan, Oguzhan; Paudyal, Sumit
    Efficient and grid-aware management of home-scale heating, ventilation, and air conditioning (HVAC) systems is one of the key enablers of demand-side management (DSM) and associated grid services in the residential sector. HVACs regulate the indoor temperature around a set point through a thermostat operating within a closed-loop control scheme. Conventional thermostats typically have a built-in temperature dead band or differential where the thermostat is idle, and HVAC stays at the most recent state (On/Off). The temperature dead band is an important control parameter that can help save energy as well as preventing frequent On/Off switching cycles leading to excessive wear and tear on the equipment. However, strategic and dynamic adjustment of the dead band can be a challenging task for an occupant. This paper proposes a mixed-integer linear program (MILP)-based tuning scheme to optimally determine the dead band. The novelty in this formulation is the inclusion of thermostat hysteresis curve modeled by piecewise techniques for tuning the dead band accurately. The proposed formulation is solved as a receding horizon manner for normal as well as under a demand response (DR) event and has been found it can achieve up to 10% reduction in energy consumption without degrading the regulation performance significantly.
  • Conference Object
    Citation - Scopus: 1
    NLP-Driven Fake News Detection: A Machine Learning Perspective
    (IEEE, 2025-05-23) Coban, Mert Korkut; Bakal, Gokhan
    The rapid spread of fake news poses a significant challenge, impacting public opinion, decision-making, and societal trust. This study explores the application of Natural Language Processing (NLP) and Machine Learning (ML) techniques for robust fake news detection. Using datasets such as ISOT Fake News, WELFake, and Football Fake News, the project employs advanced preprocessing methods and feature extraction techniques, including TF-IDF, Word2Vec, and GloVe. A comprehensive evaluation of machine learning models-Random Forest, Support Vector Machines (SVM), and Neural Networks-was conducted to identify the optimal configuration. Results demonstrate that Random Forest with TF-IDF excels in in-domain detection, achieving an F1-score of 99.70%, while Neural Networks paired with Word2Vec and GloVe embeddings outperform in cross-dataset scenarios. The study highlights the importance of dataset size, domain relevance, and feature representation in achieving high generalizability. These findings provide a scalable framework for combating misinformation on digital platforms.