I am a first year PhD student in Philipp Hennig's PN-Group. My PhD is funded through a grant by ETAS (Bosch), which offers me the opportunity to gain insight into issues of machine learning in 'real life' - applications, especially for embedded-systems. My work aims for probabilistic model-adaption in deep learning during stochastic optimization.
The choice of network architecture is an essential step in deep learning on which the performance of the model depends crucially. Most model adaption techniques are based on `outer loop' optimization (e.g. Bayesian optimization, cross-validation, etc.), which have high computational costs and are difficult to control. In my project I study ways to adapt the model directly within the `inner loop’ of the stochastic optimization algorithm that also fits the model itself. We exploit statistics (e.g. the element-covariance) of the stochastic gradients within the batch to separate relevant from irrelevant model parts.
Prior to joining the MPI, I studied Physics in Stuttgart (Germany) and spent time abroad as an exchange student at Durham University (England). I did my Master's project on diffusion of DNA-grafted colloidal particles in crowded environments, at another research group at the MPI.
Our goal is to understand the principles of Perception, Action and Learning in autonomous systems that successfully interact with complex environments and to use this understanding to design future systems