We present a practical approach to Anstreicher and Lee's masked spectral bound for maximum-entropy sampling, and we describe favorable results that we have obtained with a Branch-&-Bound algorithm based on our approach. By representing masks in factored form, we are able to easily satisfy a semidefiniteness constraint. Moreover, this representation allows us to restrict the rank of the mask as a means for attempting to practically incorporate second-order information.
Citation
IBM Research Report RC23542, IBM T.J. Watson Research Center, February 2005.
Article
View Solving Maximum-Entropy Sampling Problems Using Factored Masks