Thomas Paula
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Thomas Paula has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
Nikolaus A. Adams
3 shared publications- Alexander Bußmann
Technical University of Munich
2 shared publications - Nikolaus. A. Adams
Technical University of Munich
2 shared publications Stefan Adami
Chair of Aerodynamics and Fluid Mechanics, TUM School of Engineering and Design, Technical University of Munich, Garching bei München, 85748, Germany; Munich Institute of Integrated Materials, Energy and Process Engineering (MEP), Technical University of Munich, Garching bei München, 85748, Germany
2 shared publications- Stefan Adami
Technical University of Munich
2 shared publications Alexander Bußmann
1 shared publication
Latest Publications
- Open Access
- 31 October 2026
- Published by Elsevier in Journal of Computational Physics
- Vol. 569
A rotated-Riemann discrete-equations method for two-phase interfaces with phase transition
- 31 October 2026
- Published by Elsevier in Journal of Computational Physics
- Vol. 569
Open AccessA rotated-Riemann discrete-equations method for two-phase interfaces with phase transition
- 20 April 2026
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 11 (4)
Ray-tracing image simulations of transparent objects with complex shape and inhomogeneous refractive index
- 20 April 2026
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 11 (4)
Ray-tracing image simulations of transparent objects with complex shape and inhomogeneous refractive index
- Open Access
- 6 December 2024
- Published by Elsevier in Ultrasonics Sonochemistry
- Vol. 112, 107185
Micro-jet formation induced by the interaction of a spherical and toroidal cavitation bubble
- 6 December 2024
- Published by Elsevier in Ultrasonics Sonochemistry
- Vol. 112, 107185
Open AccessMicro-jet formation induced by the interaction of a spherical and toroidal cavitation bubble
- Open Access
L2M: Practical posterior Laplace approximation with optimization-driven second moment estimation
Uncertainty quantification for deep neural networks has recently evolved through many techniques. In this work, we revisit Laplace approximation, a classical approach for posterior approximation that is computationally attractive. However, instead of computing the curvature matrix, we show that, under some regularity conditions, the Laplace approximation can be easily constructed using the gradient second moment. This quantity is already estimated by many exponential moving average variants of Adagrad such as Adam and RMSprop, but is traditionally discarded after training. We show that our method (L2M) does not require changes in models or optimization, can be implemented in a few lines of code to yield reasonable results, and it does not require any extra computational steps besides what is already being computed by optimizers, without introducing any new hyperparameter. We hope our method can open new research directions on using quantities already computed by optimizers for uncertainty estimation in deep neural networks.
Open AccessL2M: Practical posterior Laplace approximation with optimization-driven second moment estimation
Uncertainty quantification for deep neural networks has recently evolved through many techniques. In this work, we revisit Laplace approximation, a classical approach for posterior approximation that is computationally attractive. However, instead of computing the curvature matrix, we show that, under some regularity conditions, the Laplace approximation can be easily constructed using the gradient second moment. This quantity is already estimated by many exponential moving average variants of Adagrad such as Adam and RMSprop, but is traditionally discarded after training. We show that our method (L2M) does not require changes in models or optimization, can be implemented in a few lines of code to yield reasonable results, and it does not require any extra computational steps besides what is already being computed by optimizers, without introducing any new hyperparameter. We hope our method can open new research directions on using quantities already computed by optimizers for uncertainty estimation in deep neural networks.
- Open Access
- 15 April 2019
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 4 (4) , 044003
Analysis of the early stages of liquid-water-drop explosion by numerical simulation
The early stages of the shock-driven explosion of liquid-water microdrops are studied numerically with a high-resolution discretization of the axisymmetric Euler equations. A level-set based conservative interface-interaction method is extended to allow for phase transition. The method is applied to a configuration that has been investigated in recent experiments [Stan et al., Nat. Phys. 12, 966 (2016); Stan et al., J. Phys. Chem. Lett. 7, 2055 (2016)]. The presented results show that the numerical model predicts the initial stages of the violent liquid-drop explosion dynamics accurately. Our results indicate that the deformation of the cylindrical vapor cavity within the droplet is not induced by a torus-shaped negative-pressure wave as was implied from the experimental data. We rather find that this torus-shaped wave is a shock wave, and that the observed vapor-cavity deformation is caused by interaction with a negative-pressure region preceding the torus shock. The simulation results show deviations from experimental results at later stages when the drop deformation is dominated by off-center cavitation, i.e., by effects that require extension of the underlying model to take into account generalized nucleation and recondensation.
- 15 April 2019
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 4 (4) , 044003
Open AccessAnalysis of the early stages of liquid-water-drop explosion by numerical simulation
The early stages of the shock-driven explosion of liquid-water microdrops are studied numerically with a high-resolution discretization of the axisymmetric Euler equations. A level-set based conservative interface-interaction method is extended to allow for phase transition. The method is applied to a configuration that has been investigated in recent experiments [Stan et al., Nat. Phys. 12, 966 (2016); Stan et al., J. Phys. Chem. Lett. 7, 2055 (2016)]. The presented results show that the numerical model predicts the initial stages of the violent liquid-drop explosion dynamics accurately. Our results indicate that the deformation of the cylindrical vapor cavity within the droplet is not induced by a torus-shaped negative-pressure wave as was implied from the experimental data. We rather find that this torus-shaped wave is a shock wave, and that the observed vapor-cavity deformation is caused by interaction with a negative-pressure region preceding the torus shock. The simulation results show deviations from experimental results at later stages when the drop deformation is dominated by off-center cavitation, i.e., by effects that require extension of the underlying model to take into account generalized nucleation and recondensation.