A learning theory for ranking aggregation

Joint work with Stephan Clémençon, Eric Sibony
LTCI, Télécom ParisTech, Université Paris-Saclay ; Shift Technology

We develop a statistical learning theory for ranking aggregation and assess the generalization ability of empirical ranking medians. Beyond the characterization of optimal solutions for the Kemeny aggregation problem, universal rate bounds are established and the situations where convergence occurs at an exponential rate are fully characterized.

Korba, Clémençon, Sibony. A learning theory of ranking aggregation. In Proceeding of AISTATS 2017.