METHOD FOR JOINT MODELING OF MEAN AND DISPERSION

The present invention describes a method and system for joint modeling of a mean and dispersion of data. A computing system derives a loss function taking into account distributional requirements over the data. The computing system represents separate regression functions for the mean and the dispersion as stagewise expansion forms. At this time, the stagewise expansion forms include undetermined scalar coefficients and undetermined basis functions. Then, the computing system chooses the basis functions that maximally correlate with a corresponding steepest-descent gradient direction of the loss function. The computing system obtains the scalar coefficients based on a single step of Newton iteration. The computing system completes the regression functions based on the chosen basis functions and obtained scalar coefficients.