A0893
Title: Conformal prediction for network regression
Authors: Liza Levina - University of Michigan (United States)
Ji Zhu - University of Michigan (United States)
Robert Lunde - Washington University in St Louis (United States) [presenting]
Abstract: A significant problem in network analysis is predicting a node attribute using nodal covariates and summary statistics computed from the network, such as graph embeddings or local subgraph counts. While standard regression methods may be used for prediction, statistical inference is complicated because the nodal summary statistics often exhibit a nonstandard dependence structure. Under a mild joint exchangeability assumption, conformal prediction methods that are finite-sample valid for a wide range of network covariates are shown. A form of asymptotic conditional validity is also proved that is achievable using standard nonparametric regression methods.
