Fig. 4.

The Eighty Five Percent Rule applied to the Law and Gold model of perceptual learning. a Precision of the trained network as function of training error rate. Gray dots represent the results of individual simulations—note that the empirical error rate on each run often differs slightly from the target error rate due to noise. Red dots correspond to the average precision and empirical error rate for each target error rate (error bars ± standard deviation in both measures). b Accuracy as a function of coherence for the network trained at three different error rates corresponding to near optimal (ER = 0.17), too high (ER = 0.38) and too low (ER = 0.06)