NSF Award
9108037
The general problem of nonparametric function fitting is how best to determine a smooth function which represents "noisy" data without making strong assumptions about the nature of the function beforehand. When a function of several variables is sought, both conceptual and computational problems increase in complexity; and extensions of one-dimensional techniques may or may not work, for example in applying standard estimation techniques over irregularly shaped regions or subspaces. This proposal addresses the construction and evaluation of an efficient class of estimators of characteristics of the unknown function and particularly of its integral.
