Data driven modeling is widely used for optimizing modern digital systems, addressing our inability to model these systems from first principles due to their complexity and the high dimensional data they generate. However, the data driven models often contain millions of parameters, making it difficult to understand their behavior and to trust their output. Moreover, they require large amounts of labeled data, which is not always available. There are data driven approaches targeting these challenges (e.g. surrogate modeling or unsupervised learning), but they are often limited in their application. Knowledge driven approaches, on the other hand, are often explainable and can leverage expert knowledge to deal with few labeled samples, but struggle with noisy and high dimensional data.
We are looking for candidates that would like to explore integrating data driven with knowledge driven modeling to provide trustworthy modeling of high dimensional data.
This position is funded by the InShape European project aiming to develop and demonstrate a novel metal powder bed fusion process for additive manufacturing in four different industrial use cases (aerospace, energy, space and automotive). The project has 10 industrial and academic partners. Our focus in the project is on developing data driven methods for high-dimensional sequential and/or image data for optical systems (laser beam shaping) and the monitoring, control and optimization of the manufacturing process, in order to improve product quality and sustainability of the process.
In the context of laser beam shaping, one possible application for the data driven methods will be the phase retrieval problem, where the goal is to identify the phase-distribution that is required to generate a specific irradiance profile (or point-spread function). In general, this inverse problem has no exact solution. While approximate solutions exist, they tend to be computationally expensive and/or unreliable.
The successful candidate for the PhD position is expected to:
- Perform scientific research on hybrid data/knowledge driven modelling in general and to validate the results in the InShape project;
- Publish results at (international) conferences;
- Collaborate with other group and faculty members;
- Collaborate with selected InShape project partners, attend project meetings and contribute to deliverables and project outcome;
- Assist with educational tasks (e.g. supervise(under)graduate students and lab/course assignments), max 10% of the time.