Figure 1
Visual discrimination task and training paradigm with heterogeneous stimulus similarity.
(A) Sample CS+ with exchangeable CS− stimuli for discriminative trials. The difficulty in discrimination depends on the degree of structural similarity (SSIM) between stimuli, indicated on the top. (B) Scheme of the visual discrimination swimming task: two monitors facing the ends of the arms of a Y-maze simultaneously display the positive (CS+, reinforced) and a negative (CS−, non-reinforced) stimuli (100% contrast). A submerged transparent platform below the CS+ serves as the unconditioned stimulus (US). The position of both the platform and CS+ in either arm varies pseudo-randomly over consecutive trials. During training, mice are released into the pool from a release chute and they learn to swim towards the CS+ (correct choice) in order to reach the platform and escape from the water. (C) Flowchart of a ‘training unit’ where the mice are presented with a given pair of CS+/CS−, a presentation that can be repeated up to 5 times if the mouse makes incorrect choices. (D) Color matrices showing similarity comparisons across all combinations of image pairs. Comparisons for the first set of stimuli are on the left (1–9): SSIM, 2D-xcorr, SNR, Weighted SNR, Peak SNR, MSE, NQM, VIF and VIFP. The SSIM-matrix for 1112 hybrid images is displayed on the right. Color-bar on the right; red indicates greater and blue lower similarity between images. (E–F) Standard deviation against average SSIM for each reference image with respect to the remaining stimuli, with the training CS+ depicted as a black dot. Note the well distributed similarities spanning into a ‘wide’ (E) or a ‘narrow’ range (F). CS− stimuli can be sorted by increasing (blue dots) or decreasing (red dots) similarity relative to the CS+, used as the stimulus timeline for visual discrimination training. The wide and narrow SSIM training regimes only differ in the sorting of equiprobable stimuli. The luminance distributions on the right (Set1: μ ± σ: 85 ± 1 lux, n = 300; Set2: μ ± σ: 84 ± 1 lux, n = 300) confirm that the stimuli can be treated as equiluminant.
