Doctoral Researcher in Lvlm Hallucination Detection

**About us** The University of Luxembourg is an international research university with a distinctly multilingual and interdisciplinary character. The Interdisciplinary Centre for Security, Reliability and Trust (SnT) at the University of Luxembourg is a leading international research and innovation centre in secure, reliable and trustworthy ICT systems and services. We play an instrumental role in Europe by fueling innovation through research partnerships with industry, boosting R&D investments leading to economic growth, and attracting highly qualified talent. We look for researchers from diverse academic backgrounds to contribute to our projects in areas such as: Network Security, Information Assurance, Model-driven Security, Cloud Computing, Cryptography, Satellite Systems, Vehicular Networks, and ICT Services & Applications. **Your role** Large vision-language models (LVLMs) can describe driving scenes and support decisions [Li25], but they sometimes hallucinate objects, relations, or events that are not present [Liu24,Liu25]. In a safety-critical domain, reducing hallucinations and improving robustness and trustworthiness are essential. This PhD targets principled ways to detect, analyze, and mitigate hallucinations in video-based LVLMs for autonomous driving. Objectives - Design, develop, and evaluate novel method(s) to detect and localize hallucinations in LVLM outputs for autonomous driving tasks - Investigate and propose mitigation strategies to reduce hallucinations or improve model confidence - Evaluate and benchmark against state-of-the-art methods the reliability and robustness of the hallucination detection system under diverse visual and textual conditions Research Environment The PhD student will join the Secure and Reliable Software Engineering and Decision-Making (SerVal) group at the University of Luxembourg. SerVal conducts research in security and reliability in software engineering, with a particular focus on AI, data science, and decision-making, as well as designing, testing, and debugging to improve software quality across domains such as FinTech, energy, and Industry 4.0. Within this context, the PhD will contribute to the group's growing research on trustworthy AI, focusing on methods that improve the transparency, reliability, and interpretability of intelligent systems in safety-critical domains like autonomous driving. **Your profile** - Master's degree in Computer Science, Artificial Intelligence, Robotics, or related field - Strong background in machine learning, deep learning, or computer vision - Experience with Python and common AI frameworks (PyTorch, TensorFlow) - Interest in hallucination detection, robustness, trustworthiness, and (optionally) explainability for autonomous systems - (Optional) Experience with simulation environments (e.g., CARLA) or multimodal/LVLM models - LanguageRequiremets:_ **We offer** - Multilingual and international character. Modern institution with a personal atmosphere. Staff coming from 90 countries. Member of the "University of the Greater Region" (UniGR) - A modern and dynamic university. High-quality equipment. Close ties to the business world and to the Luxembourg labour market. A unique urban site with excellent infrastructure - A partner for society and industry. Cooperation with European institutions, innovative companies, the Financial Centre and with numerous non-academic partners such as ministries, local governments, associations, NGOs **How to apply** - Applic_ations should include:_ - Curriculum Vitae - Cover letter presenting your motivation for this doctoral thesis topic, and explaining how your qualifications and aspirations align with its academic focus - Transcript of all modules and results from university-level courses taken - General information:_ - Contract Type: Fixed Term Contract 36 Month - Work Hours: Full Time 40.0 Hours per Week - Location: Kirchberg Campus - Internal Title: Doctoral Researcher - Job Reference: UOL07842 The yearly gross salary for every PhD at the UL is EUR 41976 (full time).