Fig. 5

Comparison between conditioned gradient ascent (red) used in this paper, and simple gradient ascent (green) optimization. Left gradient intensity \(\overline{\nabla }(\mathbf {E})_s\) per iteration of a single parameter \(k_s\). Right the energy \(\mathbf E \) per iteration. The conditioned gradient ascent has faster convergence to the local maxima while keeping a smooth gradient curve (Color figure online)