Learning Dynamics for AI in Health
Dynamical systems pervade our daily life. A robot picking up objects, the flow of water in a river, the trend of cost in economics, the spread of disease, and the human brain itself are all examples of dynamical systems. Understanding how these systems evolve, how they react to inputs, and how they can be controlled, is a crucial aspect of science.
In the last decade, Artificial Intelligence and data-driven methods such as Deep Learning and (Deep) Reinforcement Learning have paved the road for new approaches for studying, analysing, understanding, and controlling dynamical systems. However, the limited generalisation and the need for a huge amount of data hinder the application of these methods to all the cases with low data regimes and real-world interaction.
In this project, we study the problem of control of dynamical systems via data-driven methods. However, instead of relying only on data, we incorporate into our learning-based approaches structural priors deriving from our knowledge of the world. This research aims at understanding which priors need to be incorporated and how to do it in practice for making further steps in controlling dynamical systems. The connection of Reinforcement Learning and Optimal Control, manifold learning, dimensionality reduction, uncertainty quantification, learning of (latent) dynamical models are some of the themes that are treated.