Figure 5.4: The logarithm of the Laplace approximation (dotted line) plotted alongside the logarithm of the true posterior for a regression neural network (solid line). The plots show the logarithms plotted along the eigenvalues of the covariance matrix for the Laplace approximation. The plots are displayed in the order of highest eigenvalue first, lowest last. In the last four plots the approximation overlays the true value. The curves are shown with an offset so that they are visible.