Figure 8

The interplay between visual discrimination and learning.

Training with varying CS⁺/CS⁻ similarity exposes the flexibility of visual discrimination performance. (A–B) Complementary scatter plots for biased trials (A) and visual discrimination performance (B) as a function of training (y-axis) and CS⁺/CS⁻ similarity (x-axis). The black regions define the similarity range for biased swims (≥2; A) or successful discrimination (B). Gray dots represent unbiased choices (A) or choices undistinguishable from chance level (B, color-bars on the right; see also Methods). (C) Average discrimination performance from the complete set of discriminative trials against the average stimulus similarity during training. The vertical bars are the S.E.M. of the discrimination performance (amplified in D), while the horizontal bars represent the SSIM-range for successful discrimination (identical to shaded regions in B). Dot-size is proportional to average efficiency (see Fig. S4 ). The plot reveals a compromise: training with decreasing similarity leads to higher levels of discrimination performance but covering a smaller similarity range for successful discrimination, as opposed to training with increasing similarity. The upper inset shows that in discriminative trials the average escape latency increases with the degree of CS⁺/CS⁻ similarity. In this task, the mice benefit from having quick reaction times, because the platform (reward) is spatially dissociated from the regions where visualization of the stimulus occurs and, as a result, speed might be favored over accuracy. Continuous lines are weighted linear regressions. (D) Despite the lower discrimination performance, training with increasing similarity presents a more precise pattern of discrimination choices.