Large language models often answer confidently even when they lack relevant knowledge. This thesis asks whether they can learn a crucial form of epistemic humility: recognizing and admitting the limits of what they know. It introduces hypothetical terms, plausible concepts screened across multiple sources and likely absent from training data, to measure how readily models invent unsupported explanations. The resulting benchmarks reveal confabulation across model families, sizes, and reasoning systems. Targeted fine-tuning reduces this tendency and improves factuality across three architectures. Finally, the thesis characterizes how learned uncertainty is represented inside models and when it shapes their responses.

Date: 31.08.2026 / 14:00 Place: A-212

This thesis proposes a cyber resilience orchestration framework that extends security orchestration beyond conventional detection and response. The framework integrates Large Language Model (LLM) agents, expert-validated playbooks, automation readiness controls, and API recommendation mechanisms to support automated and human-controlled resilience operations. Experimental evaluations demonstrate the feasibility of the proposed approach under controlled conditions.

Date: 03.09.2026 / 15:00 Place: A-108

This thesis develops grounded, explainable scene understanding for sensitive-content moderation. It introduces SenBen, a benchmark of 13,999 movie frames annotated with scene graphs and sensitivity tags, and SenBen-Score, a recall-oriented metric. A vocabulary-aware distillation framework transfers Gemini annotations to a compact 241-million-parameter Florence-2 model using suffix-based object identity, Vocabulary-Aware Recall loss, and a separate sensitivity-tag head. The resulting model outperforms evaluated vision-language models except Gemini and all tested commercial safety APIs on the composite benchmark, while running 7.6 times faster and using 16 times less GPU memory than the strongest competing local model.

Date: 01.09.2026 / 11:00 Place: A-212

SALGINTR is a physician-based digital participatory surveillance system in which physicians self-report their own weekly influenza-like illness. Running at low cost (about USD 100 for the season), it was evaluated against Türkiye's national sentinel stream over the 2025/26 season (33 weeks). Of 304 registered physicians, 248 reported at least one week, contributing 4,729 person-weeks and 497 episodes. The participatory and sentinel series agreed before the epidemic (r=0.89) and discriminated epidemic weeks (AUC=0.77), with substantial alarm concordance (κ=0.835). A low-cost physician panel tracked the season in near-real time, supporting participatory reporting as a complement to laboratory-anchored surveillance.

Date: 02.09.2026 / 14:00 Place: A-212

MicroRNAs (miRNAs) regulate gene expression post-transcriptionally through sequence-specific binding to target messenger RNAs (mRNAs). Single nucleotide variants (SNVs) can disrupt miRNA–mRNA interactions and contribute to human disease. We present miRHub, a comprehensive database that integrates miRNA–mRNA duplex information with 3′UTR SNV context using matched The Cancer Genome Atlas expression and genotyping data. The miRHub portal provides integrated views for mRNA expression, miRNA expression, and co-localizing SNVs with the corresponding miRNA–mRNA duplexes. As a use case, we report statistically significant differences in mRNA regulation between different sample types based on SNV-associated duplexes.

Date: 09.06.2026 / 13:00 Place: A-212

This thesis studies multi-view, multimodal BEV perception for centerline-centric road topology understanding in autonomous driving. It focuses on improving centerline detection within a transformer-decoder framework and examines how those gains propagate to topology reasoning. The work develops three stages: mask-based centerline prediction with directional supervision and mask-Bezier fusion, Bezier-driven decoder attention through multi-point and Bezier deformable attention, and geographically disjoint plus long-range multimodal evaluation. Experiments on OpenLane-V2 and OpenLane-V1 show strong camera-only and fused camera-LiDAR performance, supporting state-of-the-art road topology understanding under consistent protocols.

Date: 16.04.2026 / 14:00 Place: A-212

This research addresses a persistent paradox in the digital economy: While data is increasingly recognized as a strategic asset, data science projects designed to leverage its potential impact continue to suffer from high failure rates. As established in management theory, measurement is the prerequisite for improvement; without the ability to objectively assess success, organizations cannot effectively detect risks or optimize their initiatives. However, the current literature lacks a formalized, operationalizable success assessment model that accounts for distinct characteristics of data science projects and is applicable across diverse project types and contexts. To bridge this gap, this thesis develops the Data Science Projects Success Assessment Model (DS PRO-S). Adopting a Design Science Research (DSR) approach, the study constructs a holistic solution that functions as a meta-model, an instantiation toolkit, and a methodology to make project success explicit, measurable, and comparable. This architecture is supported by a rigorous mathematical formalization of measurement and evaluation, aligned with the ISO/IEC 15939 standard. By introducing evaluations at both project and phase levels and decoupling success (the achievement of objectives) from health (establishing the enabling conditions for success), DS PRO-S offers a modular and asynchronous assessment capability with operational flexibility. The applicability and usefulness of DS PRO-S were validated through expert interviews and multiple case studies.

Date: 21.01.2026 / 13:30 Place: A-212

Accurate identification of splice-altering genetic variants is critical for understanding disease mechanisms and improving clinical variant interpretation. Although deep learning–based splice prediction tools perform well for canonical splice-site variants, their ability to detect exonic splice-altering variants remains limited. This limitation is primarily due to the scarcity of experimentally validated exonic variants and model architectures optimized for canonical splice motifs rather than regulatory exonic regions. Overall, this study provides a comprehensive evaluation of current splice prediction tools, demonstrates the benefit of targeted retraining for exonic variant detection, and establishes a foundation for developing more accurate and clinically relevant splice-altering variant prediction models.

Date: 15.01.2026 / 11:00 Place: A-212

This dissertation presents ID-SDM, a computational framework utilizing Influence Diagrams to model Shared Decision-Making (SDM) based on the Three-Talk Model. By representing clinicians and patients through separate IDs, the model simulates information flow via three node operations: decision alternative transfer, chance node transfer, and preference transfer. Applied to Graves’ Disease, results show that SDM achieves the perfect-information-sharing model’s optimal decision.more efficiently than other decision models. The SDM process reaches consensus in less time with upfront information sharing from both sides. When the clinician attributes greater importance to the patient's utility criteria, the clinician's decision shifts to the perfect-information-sharing model’s optimal decision.

Date: 13.01.2026 / 13:30 Place: A-212

This thesis proposes the Data Governance Capability Maturity Model (DG-CMM), a structured assessment model based on ISO/IEC 330xx standards for evaluating organizational data governance maturity. The model examines maturity across four core process areas: Data, Organization, Strategy, and Technology. DG-CMM was developed using a Design Science Research methodology in line with Becker et al. (2009), incorporating an extensive literature review, a Modified Delphi approach with domain experts, and empirical case-based validation. The model offers organizations a standardized and actionable framework to systematically identify maturity gaps, prioritize improvements, and strengthen data-driven decision-making and strategic alignment through effective data governance practices.

Date: 16.01.2026 / 09:00 Place: A-212