on-site: seminar room 403
online: MS Teams (link via E-Mail: [email protected])
Natalie Rauter (HSU): How to predict material behavior by utilizing neural networks
Numerical simulations of materials can be challenging and time consuming when dealing with complex microstructures, different scales, and damage evolution. Here, NN can be beneficial by reducing the complexity of the model and hence, the computational costs. Examples are an improved autoencoder architecture to compute heat transfer and a novel U-Net to predict crack evolution. While the first approach is based on model order reduction, the second approach deals with the spatiotemporal extension.
Jaspar S. Ibnamar (Laboratoire d’informatique, parallélisme, reseaux, algorithms, distribués): Performance profiling on modern HPC computer systems: techniques and tools
This presentation will introduce various techniques and tools for profiling the performance of HPC applications on modern computer systems. We will begin by discussing key features of contemporary architectures, followed by an exploration of code optimizations and strategies for guiding compilers toward improved code generation. Next, we will cover performance measurement methodologies and the selection of ppropriate performance metrics. Finally, we will present performance profiling tools and demonstrate how they can help identify and resolve
performance bottlenecks.