Research News

M.S. Thesis
Barış Özcan, Adaptive System for Dynamic Handling of Concept Drift: Detection, Modeling, and Weighted Ensemble Predictions

This thesis addresses the challenge of concept drift in machine learning, where evolving data patterns reduce model relevance and performance. This research proposes a dynamic system that detects and adapts to new concepts by developing tailored models for each concept. It includes leveraging ensemble strategies and mitigating class imbalances with synthetic data. By using detection techniques based on differences between datasets and performance metrics, and different prediction techniques that take account of the concept of the datasets that will be predicted this research aims to enhance model adaptability in dynamic environments, providing a comprehensive framework to tackle concept drift.

Date: 27.12.2024 / 14:00 Place: A-212

Ph.D. Thesis
Güliz Demirezen, Cross-Session EEG-Based Mental Workload Classification Using Graph Neural Networks for Reproducible Brain-Computer Interface Applications

This thesis develops a reproducible methodology for classifying mental workload using EEG signals across multiple sessions. Guidelines for reproducible research are established and a thorough review of existing EEG-based workload classification studies is conducted to assess their reproducibility status. Graph neural networks are employed for classification. Domain adaptation with optimal transport is explored for improved generalization across sessions. Subject-specific evaluations using diverse metrics are performed to assess model performance. The outcomes aim to enhance the robustness and generalizability of mental workload classification for brain-computer interfaces and other cognitive workload applications.

Date: 20.12.2024 / 13:00 Place: A-212

M.S. Thesis
Kaan Karataş, Developing A Framework to Evaluate the Usability of Virtual and Mixed Reality Environments to Practice Model-Based Systems Engineering

This thesis aims to understand the applicability of virtual reality or mixed reality environments to perform model-based systems engineering and develop a prototype for a framework for such uses. By conducting user tests with people from systems engineering and interactive application and game development background, identifies the primary advantages and disadvantages of using these environments compared to desktop environment. The outcomes serve as a strong baseline for possible future research and established that the virtual reality or mixed reality environments can be suitable for model-based systems engineering.

Date: 26.11.2024 Place: A-212

M.S. Thesis
Ümit Eronat, A Comparative Analysis of Various 3D Mesh Optimization Algorithms for Assessing Effectiveness on Sustaining Virtual Visual Illusion

This thesis presents a method of comparing the cost-effectiveness of 3D mesh simplification algorithms using the McGurk effect, where visual and auditory cues are combined to create an illusion. The study involves designing a human head mesh, animating mouth movements, and recording certain syllable sounds to produce a virtual scene. Using this virtual scene and applying three different mesh simplification algorithms on the animated head, a user study was conducted to test and measure the effectiveness of each algorithm for each different syllable in medium and high difficulty levels. Results highlight the balance between computational efficiency and perceptual accuracy, providing insights for 3D modeling and virtual reality applications.

Date: 29.11.2024 / 10:00 Place: A-212

M.S. Thesis
Yasin Aksüt, An Analysis Of Kerberoasting Attack And Detection With Supervised Machine Learning Algorithms

Active Directory (AD) is one of the most widely used directory services today, playing a key role in organizing and managing network resources within an organization. A robust security strategy is crucial to prevent and detect AD attacks, which can be difficult to detect due to their blend in with normal network traffic. One such attack is the Kerberoasting attack, which exploits weaknesses in the Kerberos authentication protocol. To detect these attacks, supervised machine learning algorithms are being proposed. And also publicly available dataset to measure the efficiency of these algorithms for Kerberoasting attacks was created and shared.

Date: 22.11.2024 / 14:00 Place: A-108