PhD in Machine Learning for Non-Stationary and Non-Static Nano-Tomography

Original title: PhD Scholarship in Machine Learning for Non-Stationary and Non-Static Nano-Tomography - DTU Compute

Recent advances in machine learning and deep learning are opening new possibilities for computational imaging, including the reconstruction of complex 3D structures from experimental imaging data. In this fully funded PhD project at DTU Compute, you will develop novel learning-based methods for X-ray nano-tomography. As state-of-the-art synchrotron imaging systems push towards ever higher spatial resolution, even minute positioning errors, mechanical drift, and sample deformations become sources of image degradation. The central challenge is to reconstruct accurate 3D and 4D representations of samples while accounting for positional uncertainties during image acquisition and changes in sample shape over time. The project will bring together neural representations, inverse problems, geometric modeling, and computational imaging to address fundamental challenges in nano-tomography. Rather than correcting acquisition errors and sample deformations as separate preprocessing steps, you will investigate how to model and estimate the sample structure, acquisition geometry, and deformation jointly from the measured data. The position is based at DTU Compute and conducted in close collaboration with EPFL and the Paul Scherrer Institute (PSI) in Switzerland, one of the leading centers for synchrotron X-ray imaging. As a PhD student, you will work in an international research environment spanning DTU, EPFL, and PSI, combining expertise in both machine learning and computational imaging with access to experimental data and advanced imaging facilities. Research visits to Switzerland are planned as part of the project.Responsibilities

• Develop novel machine learning and deep learning methods for X-ray nano-tomography, which jointly reconstruct sample structure, acquisition geometry, and deformation fields from inconsistent projection data.
• Investigate neural field representations for 3D and 4D imaging, including models that capture changes in sample geometry over time, and explore hybrid approaches combining learning-based methods with physical models of image formation and tomographic reconstruction.
• Validate and evaluate the developed methods using experimental nano-tomography data, initially focusing on integrated circuits.
• Collaborate with researchers at DTU, PSI, and EPFL, and participate in international research visits and scientific activities.
• Publish research results in leading journals and conferences in machine learning, computer vision, and computational imaging.
• Contribute to research-based teaching and supervision of MSc and BSc student projects.
• Participate in activities within the Center for Quantification of Imaging Data from MAX IV (QIM), including workshops, training courses, and interdisciplinary research collaborations.

Qualifications You are curious and motivated by developing new machine learning methods to solve challenging problems in computational imaging. A strong background in machine learning and deep learning is particularly important. We welcome candidates from computer science, applied mathematics, physics, engineering, and related disciplines who are interested in combining modern learning methods with mathematical modeling and real-world experimental data. The candidate should be able to work independently, contribute effectively to an interdisciplinary research team, and communicate scientific ideas clearly in written and spoken English. To succeed in this position, you are expected to have:

• A master's degree in computer science, applied mathematics, physics, engineering, or a related discipline.
• A strong background in machine learning and deep learning, including experience developing, implementing, or evaluating learning-based models.
• Experience with scientific programming, preferably in Python, and familiarity with relevant machine learning frameworks.
• A solid understanding of mathematical modeling, optimization, or numerical methods.

Experience in one or more of the following areas is an advantage:

• Neural representations, implicit neural fields, or other coordinate-based representations.
• Computer vision, 3D vision, geometric modeling, or structure-from-motion.
• Inverse problems, image reconstruction, or physics-informed machine learning.
• Image registration, motion estimation, or deformable modeling.
• Image processing, image analysis, or tomographic reconstruction.
• Knowledge of X-ray imaging and tomography is beneficial but not required.

• A letter motivating the application (cover letter)
• Curriculum vitae
• Grade transcripts and BSc/MSc diploma (in English) including official description of grading scale