Tez Koleksiyonu
Bu koleksiyon için kalıcı URI
Güncel Gönderiler
Öğe Real-time machine learning anomaly detection in computer networks(Türk-Alman Üniversitesi Fen Bilimler Enstitüsü, 2024) Demir, Halit CanapIn the rapidly evolving digital world, the need for advanced security measures to protect our data has steadily increased. The growing cyber threats have made it essential to develop sophisticated Intrusion Detection Systems (IDS) that can adapt to modern network environments. In this thesis, to address this need, a system that detects malicious traffic by analyzing network traffic flows using deep learning methods is proposed. Various datasets that could be used for system development were examined, and the CICIDS2017 dataset, which stands out in terms of relevance and scope, was chosen. The CICIDS2017 dataset contains a total of 15 classes, one representing normal network traffic and the others representing different types of attacks. Training the deep learning model with a consistent and balanced dataset directly impacts system performance. Therefore, pre-processing steps such as removing missing or redundant data, eliminating irrelevant features, and balancing the number of examples in different classes were performed. Dense Neural Networks (DNN) and Random Forest (RF), methods commonly used in similar studies, were selected for the proposed model. The models developed could detect network traffic involving different types of attacks with an average accuracy of 98.5%. The main goal of this study is to detect attacks on the network. Accordingly, a version of the dataset consisting of two classes—normal network traffic and attack traffic—was created. Using this dataset, another system was developed that could detect malicious traffic with 98.8% accuracy. The systems developed in this thesis aim to detect attacks in real-time within a network. Therefore, after optimizing performance through experiments with different parameters, the models were tested in a real network environment using the NVIDIA Jetson AGX Orin embedded system. For the sustainability of the developed system, training with current network traffic and attacks is also essential. In this regard, the training of the models on the embedded system was analyzed in terms of time and performance.Öğe Unraveling the capabilities of language models in news summarization performance evaluation and comparative study(Türk-Alman Üniversitesi Fen Bilimler Enstitüsü, 2024) Odabaşı, Abdurrahman; Biricik, GökselGiven the recent introduction of multiple public Large Language Models (LLMs) andthe ongoing demand for improved Natural Language Processing tasks, particularlysummarization, this thesis provides a comprehensive benchmarking of 20 recent LLMson the news summarization task. The study systematically evaluates the capabilityand effectiveness of these models in summarizing news articles across different styles,utilizing three distinct datasets. Specifically, this study focuses on zero-shot and few-shot learning settings, employing a robust evaluation methodology that integratesautomatic metrics, human evaluation, and LLM-as-a-judge. Interestingly, includingdemonstration examples in the few-shot learning setting did not enhance models’ per-formance and, in some cases, even led to worse outcomes. This issue arises mainlydue to the poor quality of the gold summaries used as references, which hinders themodels’ learning process and negatively impacts their performance. Furthermore, ourstudy’s results highlight the exceptional performance of GPT-3.5 and GPT-4, whichgenerally dominate due to their advanced capabilities. However, among the publicmodels evaluated, certain models such as Qwen1.5-7B, SOLAR-10.7B-Instruct-v1.0,and Zephyr-7B-Beta demonstrated promising results. These models showed signifi-cant potential, positioning them as competitive alternatives to private models for thetask of news summarization.
