Performance of ASD compared to standard nonlinear optimization algorithms.

<p>The four algorithms used are Nelder-Mead nonlinear simplex, Levenberg-Marquardt gradient descent, simulated annealing, and genetic algorithms. The x-axis shows the number of individual function evaluations, while the y-axis shows the error relative to the starting point. Standard methods—especially the simplex method—are most efficient for low-dimensional problems (<i>e.g.</i>, Rosenbrock’s valley), in many cases ASD is the most efficient algorithm for high-dimensional parameter spaces (<i>e.g.</i>, the 100-dimensional version of Powell’s quartic function). For the stochastic methods (ASD, simulated annealing, and the genetic algorithm), the shaded regions show the interquartile range for 40 different random seeds.