PhD Position in Efficient Test-Time Model Adaptation in Dynamic Edge Environments

Applicants are invited for a PhD fellowship/scholarship at Graduate School of Technical Sciences, Aarhus University, Denmark, within the Electrical and Computer Engineering programme. The position is available from 01 November 2026 or later. 

Research area and project description

Applications are invited for a fully funded PhD position within the Department of Electrical and Computer Engineering at Aarhus University. The successful candidate will be integrated into the A3 Lab – Adaptive & Agentic AI, directed by Dr. Behzad Bozorgtabar, who serves as the primary supervisor. This doctoral research is co-supervised in close collaboration with Prof. Qi Zhang, offering a unique interdisciplinary research environment at the intersection of Foundation Models and Edge Intelligence.

Research Vision. Deploying models in edge environments requires navigating a fundamental conflict between model complexity and environmental volatility. Real-world edge environments remain highly dynamic: data streams are continuously subject to "domain shifts" caused by fluctuating conditions, hardware degradation, or changing physical surroundings.

Traditional AI models are often brittle under these distribution shifts, leading to unreliable outputs that can compromise safety in mission-critical applications—ranging from autonomous robotics to real-time industrial monitoring. To maintain performance without the latency penalties of cloud-based recalibration, edge AI systems must become "self-aware" and capable of autonomous evolution.

Core Research Objectives. The primary objective of this PhD is to develop a high-performance, low-latency framework for Test-Time Adaptation (TTA). This involves designing autonomous architectures capable of monitoring and maintaining the reliability of unimodal and multimodal foundation models in real-time. Key research pillars include:

• Autonomous Monitoring: Developing mechanisms to detect distribution shifts and quantify model uncertainty across heterogeneous data types.
• On-the-Fly Adaptation: Designing lightweight TTA algorithms that can recalibrate models at the edge under strict latency and computational constraints.
• Efficiency and Reliability: Balancing the trade-offs between adaptation accuracy, energy efficiency, and hard real-time execution.

The candidate will join a pioneering research group focusing on the next generation of adaptive AI, with the opportunity to publish at top-tier machine learning venues (e.g., NeurIPS, ICLR, CVPR) and validate research on state-of-the-art edge computing testbeds.

Project description

For technical reasons, you must upload a project description. Please simply copy the project description above and upload it as a PDF in the application.

Qualifications and specific competences

Applications to the PhD position must hold a master’s degree (120 ECTS) in Computer Science, Computer Engineering, Electrical Engineering, Machine Learning, or a related quantitative field.

Further qualifications:

• Technical Skills: Advanced proficiency in Python and deep learning frameworks (e.g., PyTorch).
• Core Knowledge: A strong foundation in machine learning and/or computer vision. The candidate should have a specific interest in test-time adaptation, autonomous AI systems and edge intelligence.
• Advanced Architectures & Edge AI: Familiarity with modern neural networks is required. Experience with edge-specific model compression—such as knowledge distillation, lightweight design, or parameter-efficient fine-tuning—is highly advantageous.
• Attributes: A mindset for reproducibility, open-source contribution, and the ability to work across the boundaries of algorithmic AI and practical edge systems.
• from the PhD announcement.

How to apply

Please read the full job description and apply at the university homepage

Please click on the 'Apply' button above to submit your application.

Application deadline is 15 August 2026 at 23:59 CEST.