Statistical Physics with and for AI
This research axis, building upon the specifics of TAU expertise, examines how (statistical) physics principles and questions can benefit to and from machine learning. This interaction is meant toward two types of contributions. The first one is to advance the state of the art related with the analytical or computational models (e.g., identifying computational order parameters for the glass transition problem). A second contribution builds upon statistical physics to investigate some fundamental learning limitations w.r.t. information expressed in models and available in data. A third type of contribution regards the expertise gained in designing model spaces, losses and learning protocol, used to enforce functional or syntactic properties on the considered models. These issues have been considered along Giancarlo Fissore, M. Chandorkar’s ,V. Schimmenti, F. Pezzicoli, T. Bonnaire Phds, O. Bui’s post-doc (coll. CWI; space weather)
… more to come
