About Him

Image Image

He is a team leader of Physics-Enhanced Machine Learning at Max Planck Institute (MPI-DCTS), Magdeburg, Germany, in the department led by Prof. Dr. Peter Benner. In 2018, he received his Ph.D. degree in the field of computational mathematics in his group as well. In his Ph.D. research, he developed various methodologies to construct low-dimensional dynamical models of high-dimensional dynamical models from system-theoretic perspectives.

He was awarded the prize for the best Ph.D. thesis within the Faculty of Mathematics by the Otto von Guericke University Magdeburg, Germany, and GAMM also awarded him Dr. Klaus Körper prize for his excellent Ph.D. thesis in the field of Applied Mathematics and Mechanics.

His primary research is centered around machine learning for dynamical systems, learning low-dimensional suitable embedding for high-fidelity dynamical processes. Dynamics systems often obey certain physical properties, such as Hamiltonian, conservation of mass or energy, stability. Therefore, while learning dynamics using neural networks, we desire to have neural-networks, guaranteeing these key properties. Incorporating aprior physics-based knowledge while learning dynamics results in models which are more robust to uncertainty in data. Furthermore, to accelerate engineering design processes such as optimization and control, it is important to have low-dimensional models capturing important dynamics of high-fidelity models in those low-dimensional latent variables. So, we seek to identify them using auto-encoder, and also ensure that they have desired properties.

He is currently a part of Max Planck research network on big data-driven material science (BiGmax) to examine how artificial intelligence can push new findings in material science. Particularly, we seek to only discover interetable governing equations describing dynamics, but also build surrogate models for stress fields in polyscrystalline materials using e.g., U-Net and Fourier operators that respects physical-laws.

Research Interests

  • Machine learning for dynamical models
  • Physics-informed neural network artchitecture designs
  • Discovering governing equations from noisy data
  • Dimensionality reduction and reduced-order modeling
  • Contact Details

    Linkedin profile
    ORCiD Profile

    Latest News