Learning with varying CS salience. (A) We generated stimuli with variable degrees of similarity using random numbers from a set of normal distributions with fixed mean (μ = 0.5) and variable standard deviations from 0 to 0.18, with 0.02 steps (σ = 0:0.02:0.18). (B) To simulate discriminative training, stimuli were sorted according to either increasing (gray) or decreasing (black) salience (Note that such arrangements consist of the same stimuli). The shaded region covers salience levels below an arbitrary putative threshold for learning of αmin = 0.3. (C) The asymptote of learning, λ, as presented in Eq. 3, behaves as a constant (λ ≈ λmax) for highly salient items, but drops and becomes sensitive to gradients in α as α reaches αmin. We used two salience threshold levels, namely, αmin = 0 and 0.3, which led to the left and right sigmoid curves, respectively. (D-E) Predicted learning curves for stimuli with increasing (gray) or decreasing (black) salience as arranged in (B), with αmin = 0 (D), and αmin = 0.3 (E). The differences in the learning curves (black vs. gray) are due to the arrangement of varying salience used during training. Learning curves were identical to those predicted by the standard model when similarity was held constant (thick dotted lines). Discrete, numerical solution to the equations is displayed as continuous lines for visualization purposes.