Quantitative robustness of instance ranking problems

Instance ranking problems intend to recover the ordering of instances in a data set with applications scientific, social and financial contexts. In this work, we concentrate on global robustness parametric instance terms breakdown point which measures fraction samples that need be perturbed order let estimator take unreasonable values. Existing notions do not cover so far. We propose define as sign-reversal all components causes predicted potentially completely inverted; therefore, call it order-inversal (OIBDP). will study OIBDP, based linear model, for several different carefully distinguished provide least favorable outlier configurations, characterizations sharp asymptotic upper bounds. also compute empirical OIBDPs.