Figure 8.

The success rate of the three learners in a high-entropy directed network with a bottleneck: transition probabilities among all elements are as in the environment in figure 7a, but the only element that leads to the reinforcer (‘f’) is ‘e’ (see illustration in figure 6b and related TP matrix in the electronic supplementary material, table T7). The continuous learner has a marginally significant advantage over the chaining learner, which becomes more significant when the training set is shortened (see the electronic supplementary material, figure S5).