Residual Correlation as a Diagnostic for Joint-Uncertainty Gains from GP Coregionalisation
A lightweight diagnostic is introduced, D_{\rm logdet}=-\frac{1}{2}\log\det R_{\rm res}$, which represents the idealised joint negative log-likelihood (NLL) gain from modelling a full rather than diagonal residual covariance and is computable from independent GPs alone.
Fang-Qin Zhou, J. Vanschoren
· 0 citations