Uncertainty estimation and low-dimensional representation learning
Abstract
This project will support the experiments for projects in three main directions: (1) theory of deep learning. Here, we develop new theories and try to verify this on experimental benchmark data. This includes low-dimensional representation learning. (2) Uncertainty quantification with kernel-based methods. This is based on recent advances in feature learning (possibly low-dimensional) in kernels and exploiting that information efficiently. (3) identification of dynamic systems with various deep learning-based methods such as deep state-space systems and graph neural networks.