Bringing deep learning models to various embedded devices for a multitude of applications is crucial for Bosch. Our team conducts research on state-of-the-art neural architecture search (NAS) methods that are used to enable real-world Bosch systems to leverage hardware-efficient, yet high-performing machine learning models. We observe that today’s hardware-aware NAS techniques are computationally expensive when applied to new applications or hardware targets, as they can not utilize experience from prior experiments effectively.

Thus, we are looking for a PhD student who is interested in conducting cutting-edge-research on novel methods for NAS that learn from prior experience, can generalize to novel tasks and hardware while at the same time are efficient during the search process. For this, we aim to leverage existing pre-trained foundation models (FM) such as LLMs, as well as train our own models specifically for the purpose of NAS.