Date and Place: Thursdays and hybrid (live in 32-349/online via Zoom). For detailed dates see below!
Content
In the Scientific Computing Seminar we host talks of guests and members of the SciComp team as well as students of mathematics, computer science and engineering. Everybody interested in the topics is welcome.
List of Talks
-
Thu11Nov2021
12:00Online
Dr. Max Sagebaum, Chair for Scientific Computing (SciComp), TU Kaiserslautern
Title:
Aggregated type handling in CoDiPackAbstract:
The development of AD tools focuses mostly on handling floating point types in the target language. Taping optimizations in these tools mostly focus on specific operations like matrix vector products. Aggregated types like std::complex are usually handled by specifying the AD type as a template argument. This approach provides exact results, but prevents the use of expression templates. If AD tools are extended and specialized such that aggregated types can be added to the expression framework, then this will result in reduced memory utilization and improve the timing for applications where aggregated types such as complex number or matrix vector operations are used. Such an integration requires a reformulation of the stored data per expression and a rework of the tape evaluation process. In this talk we demonstrate the overhead of unhandled aggregated types in expression templates and provide basic ingredients for a tape implementation that supports arbitrary aggregated types for which the user has implemented some type traits. Finally, we demonstrate the advantages of aggregated type handling on a synthetic benchmark case.
How to join
The talk is held online via Jitsi. You can join with the link https://jitsi.rptu.de//SciCompSeminar_02. Please follow the rules below:
- Use a chrome based browser (One member with a different browser can crash the whole meeting).
- Mute your microphone and disable your camera.
- If you have a question, raise your hand.
More information is available at https://rz.rptu.de/dienstleistungen/netz-telefonie/konferenzdienste/jitsi/.
-
Thu18Nov2021
12:00Onlne
Jan Rottmayer, TU Kaiserslautern
Title:
Flow Control via Reduced Order ModelsAbstract:
The optimization of systems governed by nonlinear partial differential equations requires solving large-scale systems of equations at each iteration throughout the design process. The reduction of model dimension and complexity is essential for the application to optimal control and design problems. This talk aims to introduce methods to construct a suitable reduced order model for an unsteady flow problem, and to apply it for optimal control. We use a POD-Galerkin model in the isentropic regime to determine the optimal control actuation for the fluid flow around a circular cylinder to mitigate vortex shedding. Our results show the successful reduction of unsteadiness in the downstream of the cylinder, and validation with a high-fidelity simulation confirms the results.
How to join
The talk is held online via Jitsi. You can join with the link https://jitsi.rptu.de//SciCompSeminar_03. Please follow the rules below:
- Use a chrome based browser (One member with a different browser can crash the whole meeting).
- Mute your microphone and disable your camera.
- If you have a question, raise your hand.
More information is available at https://rz.rptu.de/dienstleistungen/netz-telefonie/konferenzdienste/jitsi/.
-
Thu25Nov2021
12:00Online
Dr. Emre Özkaya, Chair for Scientific Computing (SciComp), TU Kaiserslautern
Title:
Robust Design Optimization Package: RoDeOAbstract:
RoDeO (Robust Design Optimization Package) is a package for simulation based global design optimization. It is specifically designed for scientific/engineering applications, in which the objective function and constraints are evaluated by computationally expensive simulations.
Main features of the RoDeO Package are:
- Surrogate model based Efficient Global Optimization (EGO) strategy.
- Data driven approach.
- Easy and efficient treatment of inequality constraints.
- Surrogate models can be trained also using sensitivity data.
How to join
The talk is held online via Jitsi. You can join with the link https://jitsi.rptu.de//SciCompSeminar_04. Please follow the rules below:
- Use a chrome based browser (One member with a different browser can crash the whole meeting).
- Mute your microphone and disable your camera.
- If you have a question, raise your hand.
More information is available at https://rz.rptu.de/dienstleistungen/netz-telefonie/konferenzdienste/jitsi/.
-
Thu02Dec2021
12:00Online
Dr. Giles Strong (CERN, University of Padova)
Title:
TomOpt: PyTorch-based Differential Muon Tomography OptimisationAbstract:
The MODE introductory article*, published earlier this year proposed an end-to-end differential pipeline for the optimisation of detector designs directly with respect to the end goal of the experiment, rather than intermediate proxy targets. The TomOpt python package is the first concrete endeavour in attempting to realise such a pipeline, and aims to allow the optimisation of detectors for the purpose of muon tomography with respect to both imaging performance and detector budget. This modular and customisable package is capable of simulating detectors which scan unknown volumes by muon radiography, using cosmic ray muons to infer the density of the material. The full simulation and reconstruction chain is made differentiable and an objective function including the goal of the apparatus as well as its cost and other factors can be specified. The derivatives of such a loss function can be back-propagated to each parameter of the detectors, which can be updated via gradient descent until an optimal configuration is reached.
*MODE (2021) Toward Machine Learning Optimization of Experimental Design, Nuclear Physics News, 31:1, 25-28, DOI: 10.1080/10619127.2021.1881364
How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/61178293734?pwd=Z0lwMXd4TTZZZ1ZSSWhjcHdUWGN4dz09 -
Thu09Dec2021
14:00Online
Prof. Mathieu Morlighem, Department of Earth Sciences, Dartmouth College, Hanover, New Hampshire, USA
Title:
The challenges of modeling the ice sheets in a changing climateAbstract:
Ice sheets are dynamic systems that are among the largest contributors to sea level rise. Observations over the last three decades have shown that the Greenland and Antarctic Ice Sheets have been losing mass at an increasing rate. How the ice sheets respond to the warming of the ocean and the atmosphere has become today one of the most urgent questions in understanding the implications of global climate change. Numerical modeling is the only effective way of addressing this problem. Yet, modeling ice sheet flow at the scale of Antarctica represents many technical challenges, and many important parameters, such as boundary conditions, remain difficult to observe directly and are poorly constrained. In this talk, we discuss some of the recent technical advances made in this field, through the use of high performance computing and inverse modeling. We also show how automatic differentiation tools, such as CoDiPack, are not only helping to better understand the physics of ice sheet flow, but also help to identify the regions that are most at risk of future change in climate conditions and should be closely monitored. These tools will eventually help us better constrain the models over the observational period and reduce the uncertainty in sea level rise projections.
How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09 -
Thu16Dec2021
12:00Online
Sharad Kakran, TU Kaiserslautern
Title:
Generalization of Neural Predictors in Neural Architecture Search, Applied to Computer VisionAbstract:
In recent years, the research topic “Neural Architecture Search” has gained a lot of popularity across the research community and industry, which mostly deals with searching for an architecture under predefined constraints for a task. Although in recent years, many sophisticated methods have been proposed to reduce the search time for exploring a search space, evaluation of so many architectures searched from a search space is computationally infeasible and this motivates the use of neural predictor which predicts the accuracy of searched architectures. However, the predictions of neural predictor strongly depend on how well it is trained which requires high quality of architecture-accuracy pairs, therefore involves a time and resource consuming process of training many sampled architectures from each new search space. Architectures, such as PSPNet, SSD, used in other computer vision tasks like semantic segmentation, object detection, share the same backbones from conventional image classification architectures, Resnet, VGG, therefore the search spaces across tasks overlap to some degree with common set of operations. Therefore we wish to investigate the question if knowledge from one task can be carried to another by using a neural predictor. A well trained neural predictor should learn a good embedding by encapsulating the information about neural architectures present in the search space such as how one operation affects another, what sets of operations result in good accuracy. We wish to evaluate how well the predictor generalises to other computer vision tasks with trained architectures sampled from the same search space.
How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09 -
Thu13Jan2022
12:00Online
Dr. Domenico Quagliarella, Italian Aerospace Research Centre (CIRA), Capua, Italy
Title:
Intrusive and non-intrusive approximation techniques for efficient robust aerodynamic shape design using risk functionsAbstract:
We introduce the use of advanced risk functions, initially conceived in financial engineering, in the formulation of robust optimization problems and illustrate their application to robust design problems of aerodynamic shapes. The focus is here on techniques for increasing the computational efficiency of robust optimization processes based on risk functions. In particular, we illustrate the approximation techniques of empirical probability distributions on which the risk functions are then calculated and discuss the differences in terms of efficiency and easiness of implementation of the intrusive and non-intrusive approximation techniques. Such techniques are then applied to robust aerodynamic design problems.
REFERENCES
[1] E. Morales, A. Bornaccioni, D. Quagliarella and R. Tognaccini, “Gradient based empirical cumulative distribution function approximation for robust aerodynamic design,” Aerospace Science and Technology, 2021, 112(5), n. 106630, (2021), Elsevier BV.
[2] D. Quagliarella and E. Iuliano, “Robust Design of a Supersonic Natural Laminar Flow Wing-Body”, IEEE Computational Intelligence Magazine, 12(4), 14-27, (2017).
[3] E. Morales and D. Quagliarella, “Risk Measures in the Context of Robust and Reliability Based Optimization,” in Vasile M. (eds) Optimization Under Uncertainty with Applications to Aerospace Engineering. Springer, Cham, (2021).
[4] E. Morales, D. Quagliarella and R. Tognaccini, “Gaussian Processes for CVaR Approximation in Robust Aerodynamic Shape Design,” in M. Vasile, D. Quagliarella (eds.), Advances in Uncertainty Quantification and Optimization Under Uncertainty with Aerospace Applications, Space Technology Proceedings 8, Springer, Cham, (to appear in 2022).
[5] E. Morales, “Optimal Energy-Driven Aircraft Design Under Uncertainty,” Ph.D. Thesis, Università degli Studi di Napoli “Federico II”, Naples, Italy (2021).How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09 -
Thu27Jan2022
12:00Online
Prof. Ralf Zimmermann, Department of Mathematics and Computer Science (IMADA), University of Southern Denmark (SDU), Denmark
Title:
How to interpolate low-rank matrix decompositions? Computing the Riemannian normal coordinates in the space of orthogonal framesAbstract:
We address the problem of computing Riemannian normal coordinates on the real, compact Stiefel manifold of orthogonal frames. The Riemannian normal coordinates are based on the so-called Riemannian exponential and the Riemannian logarithm maps and enable to transfer almost any computational procedure to the realm of the Stiefel manifold. To compute the Riemannian logarithm is to solve the (local) geodesic endpoint problem.
In this talk, we present efficient matrix-algorithms for solving the
geodesic endpoint problem on the Stiefel manifold for a one-parameter
family of Riemannian metrics. The findings are illustrated by numerical experiments. We use the Riemannian normal coordinates to construct interpolated matrix curves, where the sample data matrices stem from the thin, (truncated) singular value decomposition and the compact QR-decomposition, respectively.How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09 -
Thu03Feb2022
12:00Online
Prof. Claudia Schillings, Institute of Mathematics, University of Mannheim
Title:
A General Framework for Machine Learning based Optimization Under UncertaintyAbstract:
Approaches to decision making and learning mainly rely on optimization techniques to achieve “best” values for parameters and decision variables. In most practical settings, however, the optimization takes place in the presence of uncertainty about model correctness, data relevance, and numerous other factors that influence the resulting solutions. For complex processes modeled by nonlinear ordinary and partial differential equations, the incorporation of these uncertainties typically results in high or even infinite dimensional problems in terms of the uncertain parameters as well as the optimization variables, which in many cases are not solvable with current state of the art methods. One promising potential remedy to this issue lies in the approximation of the forward problems using novel techniques arising in uncertainty quantification and machine learning.
We propose in this talk a general framework for machine learning based optimization under uncertainty and inverse problems. Our approach replaces the complex forward model by a surrogate, e.g. a neural network, which is learned simultaneously in a one-shot sense when estimating the unknown parameters from data or solving the optimal control problem. By establishing a link to the Bayesian approach, an algorithmic framework is developed which ensures the feasibility of the parameter estimate / control w.r. to the forward model.This is joint work with Philipp Guth (U Mannheim) and Simon Weissmann (U Heidelberg).
How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09 -
Thu10Feb2022
12:00Online
Henry Jäger, TU Kaiserslautern
Title: Extending Eigen-AD for CoDiPack
Abstract:
Eigen-AD is a software which provides add-on modules for the linear algebra library Eigen. Thereby it enables different optimization options for an operator overloading AD tool, for example a better handling of expression templates or implementations of symbolically derived expressions for calculating derivatives. We want to extend Eigen-AD and define the Eigen-AD interface for CoDiPack to enable these specialized path also for CoDiPack. In addition we want to check with benchmarks if the improved code path are used.
How to join
The talk is held online via Zoom. You can join with the following link:
https://uni-kl-de.zoom.us/j/63565018552?pwd=d2VvMENEN1QvNkhlS01helUwRGhYdz09