Projects

This collaboration between MLS and BOSCH focuses on the development and application of generative models for synthetic data generation, offering innovative solutions to various challenges in the industrial sector. By harnessing advanced machine learning techniques, this project aims to create high-fidelity synthetic data that mirrors real-world industrial scenarios, thereby enhancing model training, data privacy, and cost-efficiency across industries.

Working on this project:

• Edgar Torres

In recent years, Geometric Deep Learning (GDL) has become an innovative and fruitful area of research in Machine Learning. A significant factor for its success stems from the fact that the framework of GDL enables exploiting structure in the data at a model level by leveraging symmetry and invariances, which can be used to enforce various model inductive biases. This approach has been particularly successful for data structured as graphs, such models already being employed in critical fields such as finance, computer security, and healthcare. However, having computer algorithms make decisions in high-stakes applications raises many ethical, legal, and moral concerns. In consequence, these algorithms should be trustworthy and robust. In the broader context of Deep Learning, several efforts have been to improve the models' robustness and to develop methods and tools that can aid the practitioner in explaining a model's predictions. However, these advancements can be sub-optimal, misleading, and even incompatible with GDL due to not considering the structure-informed inductive biases of the models. Therefore, the Ph.D. student’s main research will focus on the design and development of GDL methods where the robustness and trustworthiness of such models are an integral part of the research.

Working on this project:

• Andrei Manolache

SimTech (Stuttgart Center for Simulation Science) is our partner in the project "Generalization and Robustness of Learned Simulators (GRLSim)."

With this project, we aim to analyze and improve the robustness and generalization behavior of ML-based surrogate models. Classical numerical solvers often trade off accuracy and computational efficiency. Their advantage is their ability to be accurate for any given PDE as long as the discretization and other parameters are properly chosen. While ML models are efficient at prediction time, they require a substantial amount of resources during training. In order to use machine learning-based surrogate models more flexibly, it would be necessary for them to generalize to situations unseen during training. Examples of such situations are (1) new parameter values of parametric PDEs; (2) generalization to higher spatial resolutions; and (3) generalization to more time steps into the future than experienced during training. While some existing ML methods have been shown to achieve limited success in either of these three settings, we are not aware of any principled study on the generalization behavior of surrogate model classes and methods that aim to generalize with respect to two or more generalization types. We aim to close this gap with the proposed project.

Working on this project:

• Marimuthu Kalimuthu, MLS Researcher
• Jan Hagnberger, Student Assistant