Optimal One-Pass Nonparametric Estimation Under Memory Constraint

For nonparametric regression in the streaming setting, where data constantly flow and require real-time analysis, a main challenge is that are cleared from computer system once processed due to limited memory storage. We tackle by proposing novel one-pass estimator based on penalized orthogonal basis expansions developing general framework study interplay between statistical efficiency consumption of estimators. show that, proposed statistically optimal under constraint, has asymptotically minimal footprints among all estimators same estimation quality. Numerical studies demonstrate nearly as efficient its nonstreaming counterpart access historical data.