Postdoc for Physics-aware machine learning for magnetism

Large scale physical simulations are important for understanding nature and for developing new technologies for the green transition. However, simulations are limited by computational power, and thus are also the technologies and materials that can we develop to e.g. aid in the green transition.

To allow larger and more complex simulations of physical systems, we wish to do simulations where the physics only have to be solved in a few discrete points, and then use deep learning to fill in the simulation in the remaining space in a physically correct way. We want to initially attempt this for magnetic fields because magnets are used in a large number of sustainable applications and their physics is well-described and understood.

To investigate if we can make deep learning physics-informed, this postdoc position will work initially on building a deep generative model, which is able to interpolate and extrapolate a high-resolution magnetic field from a few, given measurements in a physically correct way. To achieve this, the model is not only trained on simulated magnetic field data but also on fulfilling the governing physics of magnetic fields i.e. Maxwell’s equations. The now physics-aware technique can then be transferred to additional physics subsequently by defining similar appropriate models for these. We will generate training data using our own simulation framework for magnets, called MagTense.

Responsibilities and qualifications
The postdoc will consider two complementary approaches based on deep generative models e.g. generative adversarial networks, denoising diffusion probabilistic models, variational autoencoders, and normalizing flows. In the first approach, we want to embed the governing physics during the training of the deep learning architecture, while in the second approach our physical model i.e. MagTense is directly targeted and incorporated into the learning process. The position will be conducted together with a postdoc at DTU Compute, who is similarly working on physics-aware machine learning as part of a Villum Foundation Synergy Project. Based on this close collaboration with expertise from both physics and computer science, we aim to unleash the potential of combining machine learning with physical simulations to speed up the green transition.

Qualified applicants must ideally have:

• PhD degree in computer science, physics, or similar.
• Strong background in numerical work and machine learning.
• Experience with physics or willing to learn this.
• Some experience with open source software etc.
• Desirable experience with programming languages (e.g., Python or Matlab).
• Ability to work independently, to plan and carry out complicated tasks, and to be a part of a large, dynamic group.
• Good communication skills in English, both written and spoken.

Salary and appointment terms
The appointment will be based on the collective agreement with the Danish Confederation of Professional Associations. The allowance will be agreed upon with the relevant union.

The position is for a fixed duration up to 2 years. The expected starting date is the late spring/ summer of 2023.

You can read more about career paths at DTU here .

Further information
Please contact Professor Rasmus Bjørk, +45 46 77 58 95, [email protected] .

Please do not send applications to this e-mail address, instead apply online as described below.

If you are applying from abroad, you may find useful information on working in Denmark and at DTU at DTU – Moving to Denmark .

Application
Your complete online application must be submitted no later than 28 February 2023 (Danish time).

Apply at: Postdoc for Physics-aware machine learning for magnetism

Applications must be submitted as one pdf file containing all materials to be given consideration. To apply, please open the link “Apply online,” fill in the online application form, and attach all your materials in English in one pdf file. The file must include:

• A letter motivating the application (cover letter)
• Curriculum vitae
• Grade transcripts and PhD diploma

Applications received after the deadline will not be considered.

All interested candidates irrespective of age, gender, disability, race, religion or ethnic background are encouraged to apply.

DTU Energy is focusing on functional materials and their application in sustainable energy technology. Our research areas include fuel cells, electrolysis, solar cells, magnetic refrigeration, superconductivity and thermoelectrics. Additional information about the department can be found at www.energy.dtu.dk

Technology for people
DTU develops technology for people. With our international elite research and study programmes, we are helping to create a better world and to solve the global challenges formulated in the UN’s 17 Sustainable Development Goals. Hans Christian Ørsted founded DTU in 1829 with a clear mission to develop and create value using science and engineering to benefit society. That mission lives on today. DTU has 13,400 students and 5,800 employees. We work in an international atmosphere and have an inclusive, evolving, and informal working environment. DTU has campuses in all parts of Denmark and in Greenland, and we collaborate with the best universities around the world.