We investigate the Oja sign covariance matrix (Oja SCM) for estimating partial correlations in multivariate data. The Oja SCM estimates directly a multiple of the precision matrix and is based on the concept of Oja signs, which generalise the univariate sign function and obey some form of affine equivariance property. We compare it to the classical MLE as well as to estimates based on two alternative multivariate signs: the marginal sign and the spatial sign.
|Title of host publication||Proceedings of the 1st Workshop on Information Theoretic Methods in Science and Engineering|
|Publisher||Tampere International Center for Signal Processing|
|Number of pages||6|
|Publication status||Published - 2008|