Doctoral Student in Test-Time Adaptation and Agentic AI

About the role

The Department of Electrical and Computer Engineering at Aarhus University, Denmark, is recruiting a doctoral student in the newly established A3 Lab (Adaptive and Agentic AI). The project builds machine learning systems that can safely adapt after deployment under real-world distribution shifts, combining test-time adaptation with agentic decision-making mechanisms. Supervisor: Associate Professor Behzad Bozorgtabar (behzad@ece.au.dk). Preferred start: 1 August 2026. Application deadline: 1 June 2026, 23:59 CEST.

Responsibilities

• Design test-time adaptation methods for multimodal foundation models that maintain reliability when input distributions shift after deployment.
• Build agentic components that determine when, how, and whether adaptation should occur, including fallback strategies.
• Develop uncertainty estimation, calibration, and out-of-distribution detection techniques.
• Implement feedback-driven and reward-based adaptation frameworks.
• Publish results at top-tier machine learning venues such as NeurIPS, ICML, and ICLR.
• Contribute to teaching activities for up to 20% of contracted hours.

Requirements

• Master’s degree (or equivalent such as a Danish cand.polyt. or cand.scient.) in computer science, electrical engineering, applied mathematics, or a related field.
• Strong background in deep learning or machine learning, evidenced by coursework, thesis, or published work.
• Practical programming experience in Python with PyTorch or JAX.
• Good written and spoken English.

Nice to have

• Prior work on domain adaptation, continual learning, test-time training, or uncertainty quantification.
• Experience with large language models, vision transformers, or multimodal architectures.
• Conference or journal publications in machine learning or computer vision.

How to apply

Apply through the Aarhus University PhD portal at the link above. The application must include a CV, a one-to-two-page motivation letter, degree transcripts, and the names and email addresses of two references. Submit before 1 June 2026 at 23:59 CEST.