“…Spatio-temporal predictive models fit the data by leveraging on the inductive bias associated with relational information coming from time and space of the data observations [Bai et al, 2020, Pal et al, 2021, Ruiz et al, 2020, Seo et al, 2018, Wu et al, 2019. The spatial domain is typically represented as a graph associated with, e.g., a pixel grid, a 3d mesh, a road map, or a brain network [Li et al, 2018, Shuman et al, 2013, Stankovic et al, 2020. That said, however, the term spatial has to be intended in broader terms referring to any type of functional dependence existing among sensors, well beyond those correlations coming from their physical position.…”