Behzad Bozorgtabar

A fully funded PhD position is available at Aarhus University, Denmark, in the Department of Electrical and Computer Engineering, focusing on Test-Time Adaptation and Agentic AI for Multimodal Foundation Models. The position is part of the A3 Lab – Adaptive & Agentic AI, and aims to develop robust machine learning systems capable of adapting at test time under real-world distribution shifts. Research will address the challenges faced by modern foundation models, such as vision–language and multimodal models, which often degrade after deployment due to changes in data, environment, sensors, or user behavior. The project will design methods for safe and efficient adaptation of these models post-deployment, emphasizing reliability and efficiency. Key research directions include test-time adaptation, agentic decision-making mechanisms, adaptive systems for reliability monitoring, distribution shift detection, trustworthy sample selection, and lightweight model updates. The project also explores feedback-driven and reward-based adaptation, uncertainty estimation, calibration, and out-of-distribution detection. The candidate will contribute to novel algorithms, theoretical insights, and large-scale empirical evaluations, with a strong focus on reproducibility and real-world impact. The successful candidate will join a dynamic research environment, collaborate internationally, and have opportunities to publish at leading venues such as NeurIPS, ICML, ICLR, CVPR, and ECCV. Open-source contributions and interdisciplinary collaborations are encouraged. Applicants must have a relevant Master’s degree in Electrical Engineering, Computer Science, Machine Learning, Artificial Intelligence, Mathematics, or a closely related field, with strong machine learning and/or computer vision background, solid Python programming skills, and experience with deep learning frameworks like PyTorch. Prior research experience is advantageous. The position is fully funded, with salary and terms according to the collective agreement. The application deadline is 26 February 2026, and the preferred start date is 15 May 2026. For further information, contact Associate Professor Behzad Bozorgtabar at [email protected]. To apply, submit a statement of interest, CV, academic transcripts, diplomas, and the project description as a PDF via the application link. Aarhus University values equality and diversity and encourages all qualified candidates to apply.

PhD opportunity at Aarhus University in Agentic Test-Time Adaptation for Efficient and Reliable Edge Intelligence , hosted by the Department of Electrical and Computer Engineering within the Graduate School of Technical Sciences. The project sits at the intersection of computer science , electrical engineering , machine learning , computer vision , foundation models , edge intelligence , and autonomous AI . The successful candidate will join the newly established A3 Lab – Adaptive & Agentic AI , directed by Behzad Bozorgtabar (main supervisor) and co-supervised by Qi Zhang . The research focuses on building low-latency, high-reliability test-time adaptation methods for unimodal and multimodal foundation models operating in dynamic edge environments. Research themes include autonomous monitoring of distribution shifts, uncertainty estimation, on-the-fly adaptation under strict computational constraints, and balancing adaptation accuracy with energy efficiency and real-time execution. The post highlights applications in mission-critical settings such as autonomous robotics and industrial monitoring, and mentions opportunities to publish in venues such as NeurIPS, ICML, and CVPR. Funding: The position is fully funded as a PhD fellowship/scholarship, with salary and employment terms according to the applicable collective agreement. Eligibility: Applicants should hold a master’s degree (120 ECTS) in Computer Science, Computer Engineering, Electrical Engineering, Machine Learning, or a related quantitative field. Strong Python and PyTorch skills, a solid ML/CV background, and familiarity with Transformers, advanced CNNs, knowledge distillation, lightweight architectures, or parameter-efficient fine-tuning are preferred. Interest in Test-Time Adaptation, Continual Learning, Machine Unlearning, and multimodal foundation models is especially relevant. Application: Deadline is 20 May 2026 at 23:59 CEST . Applicants must include a 1-page statement of interest, CV, and academic records. A project description must also be uploaded as a PDF by copying the provided project text. Apply through the official link before the deadline; only complete applications received on time will be considered.