Meta-learning is a set of machine learning tools to leverage past and similar experiences for rapid adaptation to new tasks. Declined for control theory, it allows for designing novel controllers and estimators that exploit the often-available data from similar systems for enhanced performance with less extensive training. In this seminar, three different declinations of meta-learning are presented: (i) meta-learning for model reference data-driven control, (ii) meta-learning for black-box optimization, and (iii) in-context learning for model-free state estimators.

The seminar is part of the Systems & Control Ph.D. Seminar Series. See the event program for further information.