Abstract:
|
Warping of functional data can be viewed as a distance-preserving operation under suitable metrics. This, when coupled with a group structure on the set of warp maps, provides a natural way to decompose the original function space into amplitude and phase components. Depending on the complexity of the sample of functions observed, the amplitude space can be finite dimensional, and can thus influence convergence rates of kernel-based estimators of regression functions with functional predictors. Numerical examples demonstrating such properties will be presented.
|