My research leverages deep learning and synthetic data to speed up physical simulations, with a specific focus on accelerating fluid dynamics simulations. Due to the non-local and non-linear nature of fluid flows, these simulations can become computationally intensive, impeding iterative design and real-time simulations. Deep learning techniques have proven instrumental in alleviating this bottleneck for targeted applications. To also ensure high accuracy in out-of-distribution cases, I believe that deep learning models must align with the underlying physics. Consequently, I work with Graph Neural Networks, which efficiently capture spatial gradients of the physical fields, and I have explored incorporating SE(3) equivariance into the models.
I completed my PhD in 2023 at Imperial College London, and since then, I have been a post-doc in Thuereys’ group.
Google scholar: https://scholar.google.com/citations?user=SsgPftwAAAAJ&hl=es
Contact
E-mail: m.lino (at) tum.de
Room: 02.13.039