Our Fields Of Expertise

Hybrid models offer a powerful approach by combining data-driven models, derived from observation or simulation, with first-principle based models rooted in statistics, physics, or chemistry. This integration proves particularly valuable in situations where neither approach alone suffices. By integrating the two into a joint architecture, hybrid modeling can combine the strengths of both approaches offering a way to enable more accurate, interpretable predictions while making the solutions more data efficient and easier to validate.

Deep Learning has delivered some of the most exciting advances in AI in recent years. However, deploying deep learning systems in the real world requires overcoming several challenges. Our research is constantly pushing boundaries of this technology to make deep learning systems in Bosch products safe, robust and efficient.​

AI systems need to understand natural language to interact with humans and extract knowledge from unstructured sources. Therefore, we are developing novel methods for modern natural language processing and knowledge representation as well as reasoning techniques to automatically structure and process information within the whole Bosch organization.​

Incorporating domain expert knowledge into neural networks' predictions and output is often challenging. Neuro-symbolic AI aims at integrating elements from both deep learning and automated reasoning, in order to achieve the best of both worlds: the flexibility and data-driven performance of deep learning with the logical consistency and grounded reasoning of symbolic methods.​​

Understanding complex processes and behaviors in the real world requires explicit representation and handling of uncertainties, and integration of formal domain knowledge. Therefore, we are developing new methods in probabilistic modeling that address these challenges, in particular for dynamical systems.​​

Many applications require more than just making predictions – we want to control or optimize. In particular, we focus on methods that efficiently learn from as little real-world trials as possible. Therefore, we are developing new methods for optimization and control, with a focus on reinforcement learning.​​