Parameter Estimation in the Presence of Bounded Data Uncertainties (Q4389072)

A new parameter estimation formulation with prior bounds on the size of the allowable corrections to the data is introduced. The solution proceeds by first providing a geometric formulation of the min-max problem formulated here. The algebraic derivation starts by showing how to reduce the min-max problem to a minimization problem. The parameter of the regularization step is shown to be obtained as the unique positive root (if it exists) of a secular equation and as a function of the given data. The global minimum of the minimization problem is found. The conclusions of earlier analysis is collected in the final theorem that contains the algorithm for the construction of the global minimizer. The parameter of the regularization is determined by the algorithm. The new method guarantees that the effect of the uncertainties will never be unnecessarily overestimated. The case in which only selected columns of the coefficient matrix are subject to permutations is also considered.