Minimax andL_{1}curve fitting in non-Gaussian MAP estimation

A class of non-Gaussian estimation problems is equivalent to minimax and L
1
curve fitting. The curve fit is shown to be algebraically dual to optimization of a positive semidefinite quadratic form with linear inequalities, which is solved by a fast quadratic program based on Graves' simplex algorithm. An example compares the performance of this estimator with the (suboptimal) minimum mean-squared error (MMSE) estimations generated by quadratic curve fitting.