Saarland University

Selected work

Representative Papers

When Feasibility of Fairness Audits Relies on Willingness to Share Data: Examining User Acceptance of Multi-Party Computation Protocols for Fairness Monitoring

Feb 02, 2026

Fairness auditing typically requires users to share sensitive data, yet user acceptance of multiparty computation (MPC) protocols designed to enable such auditing under privacy-compliant conditions remains unclear. This study addresses this gap through an online survey of 833 European participants, combining a discrete choice experiment with direct evaluation questionnaires to systematically examine user acceptance of different MPC protocol designs for fairness monitoring. For the first time, it reveals a divergence in how users weigh risk-related attributes (e.g., privacy protection) against benefit-related attributes (e.g., fairness objectives), and integrates individual privacy and fairness preferences into an acceptance model. Results show that while users prioritize privacy mechanisms in direct evaluations, they place greater emphasis on fairness outcomes in simulated choices—both factors significantly shaping willingness to adopt MPC protocols and offering critical behavioral insights for compliant deployment.

2 citationsRead paper

Robustness and Cybersecurity in the EU Artificial Intelligence Act

Feb 22, 2025

The EU AI Act (AIA) exhibits structural deficiencies—particularly ambiguous legal definitions and insufficient technical specifications—in its robustness and cybersecurity requirements for high-risk AI systems (Art. 15) and general-purpose AI models (Art. 55). This paper is the first to systematically identify and analyze the law–technology gap embedded in these provisions. Leveraging interdisciplinary analysis—including statutory interpretation, machine learning robustness theory (e.g., adversarial robustness, out-of-distribution generalization), and cybersecurity practice—we assess the operational feasibility of the requirements. Our contribution is a cross-disciplinary compliance framework that delivers actionable recommendations for the European Commission’s guidance documents, harmonized standard development, and the benchmarking methodology stipulated under AIA Art. 15(2). By aligning legal terminology with empirically grounded ML security research, the framework advances precise, implementation-ready resilience governance—thereby addressing a critical gap in the AIA’s regulatory architecture.

1 citations1 influentialRead paper

Bowling with ChatGPT: On the Evolving User Interactions with Conversational AI Systems

Feb 01, 2026

This study investigates the evolving dynamics of user interactions with large language model–driven conversational AI systems, focusing on shifts in interactional intent, social framing, and guidance patterns. Leveraging 825,000 real-world ChatGPT dialogues donated by 300 users under GDPR data rights, the research combines quantitative content analysis with conversational trajectory tracking to reveal three key trends marking the transition of conversational AI from a functional tool to a social partner: expansion into sensitive domains such as health and mental well-being, increasing socialization of interactions, and a marked rise in model-led guidance. Notably, following the release of GPT-4o, model-dominated dialogues increased fourfold, accompanied by heightened system anthropomorphism and growing user emotional reliance, underscoring a profound transformation in human–AI relational dynamics.

1 citationsRead paper

SL-CBM: Enhancing Concept Bottleneck Models with Semantic Locality for Better Interpretability

Jan 19, 2026

This work addresses the limited spatial locality in existing Concept Bottleneck Models (CBMs), which hinders precise alignment between concepts and semantically meaningful image regions, thereby undermining interpretability credibility. To overcome this limitation, the authors introduce, for the first time in CBMs, a combination of 1×1 convolutions and cross-attention mechanisms to generate faithful saliency maps tightly coupled with the model’s reasoning process. They further employ contrastive and entropy regularization to jointly optimize prediction accuracy, map sparsity, and explanation fidelity. Extensive experiments demonstrate that the proposed approach significantly improves local concept-region alignment, enhances explanation clarity, and boosts intervention effectiveness across multiple image datasets, all while maintaining competitive classification performance.

1 citationsRead paper

Deterministic Negative-Weight Shortest Paths in Nearly Linear Time via Path Covers

Nov 11, 2025

For the single-source shortest paths (SSSP) problem with negative edge weights and negative cycle detection in directed graphs, all prior near-linear-time algorithms rely on low-diameter decompositions and are randomized. This paper presents the first deterministic near-linear-time algorithm, achieving a time complexity of $ ilde{O}(m log(nW))$, which matches the optimal bound for deterministic SSSP in such graphs. The key innovation is the introduction of *path covering*—a novel structural primitive—that enables the first complete derandomization of low-diameter-decomposition-based approaches. Leveraging the integrality of edge weights and an efficient path-covering construction, our method avoids traditional random sampling entirely. This resolves a long-standing open problem in deterministic graph algorithm design and provides a scalable new tool for optimization on directed graphs.

1 citationsRead paper

Recent publications