Ghost Image Processing for Minimum Covariance Determinants

In this paper we describe a ghost image processing (see Glover [Glover, F. 1994. Optimization by Ghost image processing in neural networks. Comput. Oper. Res. 21 (8) 801--822.] application to the problem of finding the minimum covariance determinant (MCD) estimator of multi-variate shape and location (see Rousseeuw [Rousseeuw, P. J. 1985. Multivariate estimation with high breakdown point. W. Grossman, G. Pflug, I. Vincze, W. Werz, eds. Mathematical Statistics and Applications , Vol. B. Dordrecht, Reidel.]). The MCD is resistant to contamination and has other desirable statistical properties but is difficult to compute. Ghost image processing offers an opportunity to make use of knowledge of the form of solutions when constructing algorithms to solve hard combinatorial optimization problems. Experimental results and comparisons with steepest descent lend additional insights. INFORMS Journal on Computing , ISSN 1091-9856, was published as ORSA Journal on Computing from 1989 to 1995 under ISSN 0899-1499.