Abstract :
A new weighted mean-square-error (MSE) procedure for pattern classification is introduced. The method iteratively repeats the misclassified samples. Three theorems on redundancy and the least square generalized inverse solution for an inconsistent set of equations are presented and proved. The resulting algorithm is presented together with a convergence proof for the linearly separable case. Several examples are included that demonstrate the advantage of the method over the MSE solution for both the separable and nonseparable cases.
Keywords :
Algorithm, design set, discriminant function, generalized inverse, mean-square error, pattern classification, redundancy, weight vector.; Cost function; Equations; Iterative algorithms; Least squares approximation; Least squares methods; Pattern classification; Pattern recognition; Redundancy; Relaxation methods; Vectors; Algorithm, design set, discriminant function, generalized inverse, mean-square error, pattern classification, redundancy, weight vector.;
