The network learns a predictive model mimicking [31].

Similar to [31], we stimulated the network either with the sequence “ABCD” (experiment) or with all permutations of this sequence (control). After a period of self-organization, the network was tested with the sequences “ABCD” and “DCBA”. (a) Mean firing rates (“sequence magnitude”) for both test sequences. (b) For the experimental condition, the network was also tested with the sequences “ABCD”, “A_CD”, and “E_CD”. Error bars represent SEM over 20 independent realizations. ⋆ indicates p < 0.05 for a two-sided t-test assuming independent samples and identical variances.