Varying the size of the association layer.

<p><b><i>A</i></b>, Scaling with unchanged learning parameters <i>β</i> and <i>λ</i>. Left, convergence rate (proportion of 100 networks that learned the saccade/antisaccade task). Error bars denote 95% confidence intervals. Right, median convergence speed (number of trials to criterion). <b><i>B</i></b>, Left, convergence rates with adjusted learning parameters. Bar shading indicates parameter setting (see legend in right panel). Right, median convergence speed with optimized parameters.