Kernel machines with missing responses (Q2209830)

The paper focuses on the inference problem for missing data. The main result is the doubly-robust kernel-machine estimator. For review of the doubly-robust estimators readers may refer to work of \textit{H. Bang} and \textit{J. M. Robins} [Biometrics 61, No. 4, 962--972 (2005; Zbl 1087.62121)]. The main challenge here is that the estimation must converge uniformly over a set of functions that grows with the sample size, moreover, the augmentation term is supposed to ensure both doubly-robustness and convexity. The consistency is proved and convergence rates for quadratic-loss doubly-robust kernel machines are calculated. The efficiency of the proposed kernel-machine families is illustrated on regression and classification problems.