Fig. 3: Dimensionality reduction also contributes to transfer effects.
From: Abstract rule learning promotes cognitive flexibility in complex environments across species
a Faster rule learning with prior experience. Both the learning trial (left) and the performance change point (right) occurred earlier in experienced rats learning the task rules alternate/go-silent as compared to naïve animals (two-sided Mann–Whitney tests). Replication of RL model findings (as in Fig. 2d) for rats learning alternate (N = 19; b) and go-silent (N = 19; c) as a first rule. Dashed lines connect values (circles) of the same rat; filled circles represent the best (i.e., highest cross-validated likelihood) model of each rat. Model comparisons between ACL and all other models are Benjamini–Hochberg corrected two-sided Wilcoxon matched pairs test with *p < 0.05, **p < 0.01, ***p < 10−3. See Supplementary Fig. 6b, c for RL models with continuous attention scores. d Scatter plot showing results from the regression model using data from naïve and experienced rats learning the rules alternating and go-silent. The learning trial was predicted (R2adjusted) by median attention-at-choice (p = 1.6 × 10−5) but not median attention-at-reward (p = 0.86). A variable coding for learning experience was also significant (p = 4 × 10−3). Each dot represents the empirical vs. predicted learning trial. e In the absence of a task rule, strategies are selected earlier if they have been previously reinforced (i.e., in experienced rats/fourth rule) as compared to a naïve cohort (first rule), two-sided Mann–Whitney test. f Perseveration on the previously reinforced strategy alternate predicts when the rule go-silent is learned, indicating a negative transfer effect (each dot corresponds to one rat, Pearson correlation). g The number of times a rat used the strategy alternate in previous rules predicted how fast a rat would learn the task rule alternate (Pearson correlation). Box plots showing median, 25%–75% percentile, whiskers: 1.5 IQR, and outliers. Created in BioRender. Böhme, N. (2025) https://BioRender.com/b41u710. Source data are provided as a Source Data file.
