Robust Non-Parametric Function Fitting
提出一种稳健的非参数函数拟合方法，结合M估计与核回归，证明其一致性与渐近正态性，且满足极小化极大性质，适用于误差分布存在污染的情况。
SUMMARY A robust non-parametric function fitting method is introduced. The estimate is motivated from the theory of M-estimation and of kernel estimation of regression functions. Consistency and asymptotic normality are shown. Bias and variance rates are the same as those previously obtained by Gasser and Müller (1979) for linear smoothers. The estimate satisfies a minimax property, i.e. it minimizes the maximal asymptotic variance as the error distributions vary over a suitable contamination neighbourhood.