On nonparametric curve estimation with compressed data

Modified kernel estimators calculated from compressed data for density estimation and signal recovery problems are proposed. An asymptotically optimal compression technique utilizing the quantile process and data binning is employed. The statistical accuracy of the introduced kernel estimators is studied, i.e., we derive mean squared error results for the closeness of the these estimators to both the true functions and the kernel estimators determined from non-compressed data.