Fig. 4
From: Dissociating task acquisition from expression during learning reveals latent knowledge
Performance variance across rodents arises from contextual factors. a Left: normalized discrimination of all mice in the auditory go/no-go task over the over the course of learning (N = 7 mice) in the reinforced context; right: same as left but in the probe context (N = 7 mice). b Number of trials required for individual animals to reach expert performance levels (d’ > 2.0, false-alarm rate <50% for at least 100 trials) in the reinforced (black) and probe (gray) contexts. The number of trials required for expert performance is significantly more stereotyped across animals in the probe context than in the reinforced context (probe: st.d. = 358 trials, range = 1348–2332, n = 7 animals; reinforced: st.d. = 2047 trials, range = 2187–8838 trials, N = 7 mice, F(6,6) = 32.63, p = 5.025 × 10−4, two-tailed two-sample F-test for equal variances). c Left: normalized discrimination of rats in the Pavlovian feature-negative discrimination task over the over the course of learning (N = 6 rats) in the reinforced context; right: same as left but in the probe context (N = 6 rats). d Number of trials required for individual animals to reach expert performance levels (discrimination >25% for at least 1 day of training, i.e., 16 trials) in the reinforced (black) and probe (gray) contexts. The number of trials required for expert performance is significantly more stereotyped across animals in the probe context than in the reinforced context (probe: st.d. = 0.51 days, range = 2–3 days, N = 6 rats; reinforced: st.d. = 1.72 days, range = 3–8 days, F(5,5) = 11.12, p = 0.0194, two-tailed two-sample F-test for equal variances). e Schematic of network model with colors indicating parameter groups. f Left: Percentage of inter-individual variation for mice performing the auditory go/no-go task explained by the four core model parameters: inhibitory scaling: 99.7%; learning rates: 74.8%; initial weights: 70.8%; noise: 54.6%. Right: same as left but for modeling of rats learning the Pavlovian feature-negative discrimination task (variation explained by inhibitory scaling: 80.7%; learning rates: 53.0%; initial weights: 65.2%; noise: 60.6%)
