Deep neural network models of high dimensional data in a supervised learning setting can reach very high performance. Nevertheless, their applicability is still challenged by a number of drawbacks; their behavior is hard to interpret, it is hard to provide strong guarantees of generalization, it is hard to incorporate existing domain knowledge in their training process, and it is hard to apply in real contexts where data is noisy, corrupted or violating the i.i.d assumption.
Several topics related to these challenges are of specific interest for this Postdoc positions. Specifically, we are looking for candidates with an interest in studying:
- Disentanglement of factors for high dimensional data, particularly sequential and time series
- Stability in Deep Learning, modeling highly sensitive systems (i.e. which exhibit chaotic behavior)
- Generative models for simulation of complex systems (e.g. physical processes, optimal control)
- Geometric DL models of non-Euclidian data (e.g. graphs and manifolds)
- Deep Metric Learning for capturing expert knowledge via psychometric measurements
- Reliable machine learning in the presence of noisy, missing or corrupted data.
As part of your application portfolio, you are highly encouraged to submit a research statement on one of the listed topics or a topic that you believe will address the challenges stated above.
This position is funded by the MADEin4 European project targetting the design and validation of data driven methods and tools for metrology in semiconductor industry. The project is joined by ca. 40 industrial and academic partners from Europe.
The successful candidates are expected to:
- perform scientific research in the domain described
- publish results at (international) conferences
- collaborate with other group and faculty members
- collaborate with selected MADEin4 project partners, attend project meetings and report on results
- assist with educational tasks (e.g. supervise Master students and internships)