Research group in data-driven fluid mechanics led by Luca Magri.
- ESN - Validation - Validation and optimization of Echo State Networks.
Robust Optimization and Validation of Echo State Networks for learning chaotic dynamics. A. Racca and L. Magri, Neural Networks (2021). - ESN for extreme events - Control and prediction of extreme events in turbulent shear flow with Echo State Networks.
Data-driven prediction and control of extreme events in a chaotic flow. A. Racca and L. Magri, Physical Review Fluids (2022);
Statistical prediction of extreme events from small datasets, A.Racca and L.Magri, Lecture Notes in Computer Science (2022). - PISR - Physics-Informed Super Resolution.
Physics-Informed CNNs for Super-Resolution of Sparse Observations on Dynamical Systems. D. Kelshaw, G. Rigas and L. Magri,
NeurIPS Workshop on Machine Learning for the Physical Sciences (2022). - PICR - Physics-Informed Corruption Removal.
Physics-Informed Convolutional Neural Networks for Corruption Removal on Dynamical Systems. D. Kelshaw and L. Magri,
NeurIPS Workshop on Machine Learning for the Physical Sciences (2022). - HCTA - Hard-constrained neural networks for thermoacoustics.
Hard-constrained neural networks for modeling nonlinear acoustics. D.E. Ozan and L. Magri, Physical Review Fluids (2023). - Adjoint-ESN - Data-driven inference of adjoint sensitivities.
Adjoint Sensitivities of Chaotic Flows without Adjoint Solvers: A Data-Driven Approach. D.E. Ozan and L. Magri, Lecture Notes in Computer Science (2024);
Data-driven computation of adjoint sensitivities without adjoint solvers: An application to thermoacoustics. D.E. Ozan and L. Magri, Physical Review Fluids (2024).
- VKI-ULB lecture series - Demonstration of echo state network (ESN) and long short-term memory network (LSTM) created for the VKI lecture series: Machine Learning for Fluid Mechanics 2024.
- NewtonWorkshop2023 - Demonstration of super resolution and thermoacoustic neural networks created for the Newton Institute Tutorial 2023.
- KolSol - pseudospectral Kolmogorov flow solver, contains both NumPy and PyTorch implementations to allow for autograd-compatible workflows.
