A1065
Title: A conformal prediction approach to predict populations of networks
Authors: Matteo Fontana - Royal Holloway, University of London (United Kingdom) [presenting]
Anna Calissano - University College London (United Kingdom)
Simone Vantini - Politecnico di Milano (Italy)
Gianluca Zeni - Politecnico di Milano (Italy)
Abstract: Despite the growing importance of population of network data and its analysis, a significant gap exists in the literature with respect to predicting these objects and quantifying the certainty of those predictions. To address this, a new technique is proposed for generating statistically valid prediction sets for populations of networks. The method, rooted in conformal prediction, uniquely handles both labelled (fixed-node) and unlabeled (variable-node) graph structures, the latter by defining sets in a discrete quotient space. This approach offers three key advantages: It requires no distributional assumptions, provides finite-sample guarantees, and yields results that are easily interpretable. Simulation studies confirm the method's theoretical properties, while an analysis of player passing networks from the FIFA 2018 World Cup showcases its real-world applicability.
