A0906
Title: Conformal prediction for dyadic regression
Authors: Robert Lunde - Washington University in St Louis (United States) [presenting]
Ji Zhu - University of Michigan (United States)
Liza Levina - University of Michigan (United States)
Abstract: Dyadic regression, which involves modeling a relational matrix given covariate information, is an important task in statistical network analysis. Uncertainty quantification is considered for dyadic regression models using conformal prediction. Finite-sample validity of the procedures is established for various sampling mechanisms under a joint exchangeability assumption. The proof uses new results related to the validity of conformal prediction beyond exchangeability, which may be of independent interest. It is also shown that, under certain conditions, it is possible to construct asymptotically valid prediction sets for a missing entry under a structured missingness assumption.
