Extended Data Fig. 6: Visual recognition memory becomes exemplar-specific after extended training.

a, Stimuli in the test2 session (see timeline in Fig. 1b). b, Anticipatory licking behavior in test2 (n=5 mice). c, Example projections of neural data onto the coding direction of leaf1 vs leaf2. d, Similarity index (from Fig. 2i) of leaf3 and circle1 stimuli for the leaf1 vs leaf2 coding direction (task mice: n=5 mice; unsupervised: n=6 mice; naive: n=7 mice; unsupervised grating: n=3 mice, 5 sessions). e, Schematic of observed ‘de-orthogonalization’ effect, where an initially symmetric projection of leaf3 becomes asymmetric and more similar to the leaf2 neural vector. Neural vectors are referenced with respect to the center of the leaf1-leaf2 axis. f, Stimuli in the test3 session (see timeline in Fig. 1b). g, Anticipatory licking behavior in test3 (n= 5 sessions in 3 mice). h, Example leaf1-selective neurons (medial region) during leaf1 and swap trials, sorted by responses in the leaf1 corridor for naive, task and unsupervised mice. Also shown (right) are the responses on swap trials after reversing the swap manually (‘unswapped’). i, Average correlation in position preference between leaf1 and the swap as well as the unswapped responses, shown for all three groups of mice across regions (task mice: n=3 mice, 5 sessions; unsupervised: n=4 mice, 8 sessions; naive: n=3 mice, 10 sessions). All data are mean ± s.e.m. *P <0.05, **P <0.01, ***P <0.001.