Fast Adaptation Methods for Continual Optimal Control

The main research question is to set the foundations of fast adaptation in continual learning. His hypothesis is that this problem can be solved by capturing information-rich invariant representations using tools of causal inference and intervening into the environment to maximize information gain. His project will address three fundamental questions:

  • Which exact mechanisms should be used to capture invariance?
  • How can optimal interventions be planned?
  • How can it be theoretically guaranteed that the eventual feedback loop constructs a virtuous cycle?
The project is funded by the Novo Nordisk Foundation.