Adjustment of learning to reward probability in the superior metaplastic models.

<p>(<b>a</b>) Plotted are the effective learning rates for potentiation () and depression () events as a function of <i>p</i>_<i>r</i>. The effective learning rate on potentiation (depression) events increases (decreases) as reward probability increases, with a crossover at <i>p</i>_<i>r</i> = 0.5. (<b>b</b>) Signal in superior metaplastic models. (<b>c</b>) Matching of the sensitivity to noise in the metaplastic models. Plotted are the normalized sensitivity (d<i>S</i>/d<i>p</i>_<i>r</i>, denoted as (Δ<i>S</i>)_<i>norm</i>) and one-step noise (<i>η</i>) as a function of <i>p</i>_<i>r</i> for three examples of superior metaplastic models with different numbers of meta-states. The sensitivity profile better matches the noise profile as <i>N</i> increases.