Probabilistic principles for latent space exploration in deep learning

This research project is concerned with combining probabilistic principles with neural network models, in order to improve their interpretability, robustness, and reliability in real-world applications. Probabilistic methods can also help build models that know when they don't know, and are capable of quantifying uncertainties related to their predictions. The project has two parts: The first is related to model building and design by allowing specification of a priori knowledge in the model structure. The second part is related inference and learning in the hidden feature space of the model in applications, where the model exhibits dynamical behaviour.