Data sharpening is a technique for perturbing data in certain optimal ways in order that when nonparametric methods are applied, the results are improved according to certain criteria. For example, data sharpening can be used to reduce bias in nonparametric regression. A form of constrained data sharpening can ensure monontonicity in nonparametric regression or unimodality in density estimation.
Some researchers at the UBC's SIMLAB (Data Visualization Laboratory) are working on new forms of data sharpening. A penalized form of data sharpening is under study as are new optimization methods.