This study presents a mobile authentication method based on modeling the distribution of touch behavior features. Instead of using summary statistics, the approach represents user interactions as probability distributions and compares them using KL divergence. A Siamese neural network is used to learn differences between users. The method is evaluated on a public dataset, showing improvements over baseline models in terms of error rates. Results suggest that distribution-based modeling can provide useful information for continuous authentication without requiring active user input.

Date: 21.08.2025 / 10:00 Place: A-212

This thesis investigates adaptive window sampling strategies for continuous mobile behavioral biometric authentication using accelerometer data. A dual-branch Siamese model architecture leveraging time-series and frequency-domain features is proposed to generate discriminative embeddings. Multiple triplet selection techniques—including session-balanced, session-weighted, entropy-filtered, and position-aware sampling—are explored to enhance model generalization. A window-ensembled inference strategy is employed to improve verification robustness. Experimental results demonstrate the effectiveness of targeted sampling and highlight the importance of temporally informed triplet construction in mobile biometric systems.

Date: 21.08.2025 / 11:00 Place: A-212

This study investigates how sequence variations in microexon and intronic regions influence splicing outcomes under differential SRRM4 expression. A custom CNN model was developed to predict PSI values from sequence data, followed by interpretation via DeepLIFT and motif discovery using TF-MoDISco-lite. The approach uncovered cis-regulatory patterns predictive of SRRM4-sensitive splicing, providing insights into the regulatory logic governing microexon inclusion.

Date: 01.09.2025 / 11:00 Place: A-212

This thesis proposes a new analysis method aimed at identifying the root causes of problems encountered in the business world. Combining the qualitative management tool Relationship Diagram with the quantitative modeling tool Bayesian Networks, this method offers a probabilistic analysis process that takes into account both expert opinions and data containing uncertainty. This approach, which incorporates the human factor into the process, This approach, which incorporates the human factor into the process, is adaptable not only to a specific sector but also to different fields. The developed method has been tested with a real case study in the field of electronic component production and has contributed to a deeper understanding of business processes.

Date: 29.07.2025 / 10:30 Place: A-212

This thesis investigates the potential of EEG-based features to distinguish between Attention Deficit Hyperactivity Disorder (ADHD) and Major Depressive Disorder (MDD), two conditions that often share overlapping symptoms and neurophysiological characteristics. Focusing on power spectral density (PSD) across frequency bands and mean curve length (MCL) as a complexity measure, the study applies multilayer perceptron (MLP) classifiers to a clinically diverse sample under both eyes open and eyes closed resting state conditions.

Date: 16.07.2025 / 14:00 Place: A-212

In energy and latency-constrained embedded systems, binary neural networks offer dramatic reductions in compute and memory by binarizing weights and activations. This thesis focuses binarizing a lightweight neural network model for classification. Using MNIST, CIFAR-10, and ImageNet, it compares full-precision models with binarized models according to accuracy, resource use, and performance trade-offs. Moreover, this thesis surveys AI hardware frameworks for FPGA-based reconfigurable systems in terms of compatibility with trending technologies and optimization strategies as a comprehensive guide.

Date: 08.07.2025 / 10:00 Place: B-116

This thesis investigates the demographic history of Anatolian sheep. Sheep are among the most important livestock species, supplying meat, milk and wool. Analysing new genome data from ancient and present-day Anatolian sheep reveals a substantial eastern-related genetic influx after the Neolithic and uncovers a population structure in present-day Turkish breeds defined by geography and tail morphology. Fat-tailed breeds, mainly from eastern Türkiye, show greater genetic affinity to Iron Age Anatolian sheep than thin-tailed breeds do. Conversely, thin-tailed breeds, from western Türkiye, retain more Neolithic-like ancestry than fat-tailed breeds. These findings show how Anatolian sheep gene pool shaped over thousands of years.

Date: 07.07.2025 / 11:00 Place: A-212

This thesis proposes a transformer-based framework with post-inference strategies designed to improve temporal coherence and semantic consistency in range-based, multiclass anomaly localization for multivariate time series.

Date: 25.06.2025 / 09:00 Place: A-212

As next-generation avionic platforms become increasingly connected, systems that were once isolated are now exposed to cyber threats. Communication standards like MIL-STD-1553, widely used in commercial, military, and aerospace platforms, were originally designed without security considerations, resulting in growing vulnerabilities. Implementing conventional security upgrades is costly and brings certification challenges. Intrusion Detection Systems (IDS) offer a non-intrusive alternative, requiring no hardware or software changes. This study aims to enhance the security of MIL-STD-1553 communication buses by integrating a hardware fingerprinting-based IDS and evaluates the effectiveness of machine and deep learning methods in detecting unauthorized devices on the bus.

Date: 02.06.2025 / 11:00 Place: A-212

This thesis aims to reduce the need for pixel-level labeled data for semantic segmentation. A DeepLabV3+ based model trained on synthetic images is supplemented with a Domain-Adversarial Neural Network (DANN), an adversarial domain adaptation method, to adapt to real images. The model is applied in unsupervised and semi-supervised domain adaptation scenarios. In the semi-supervised adaptation method in particular, similar performance was achieved using 92% less labeled real data compared to the DeepLabV3+ method trained with supervised learning and without domain adaptation. This study provides an effective solution that reduces the burden of image labeling.

Date: 26.05.2025 / 15:00 Place: A-212