Deviance detection in our model.

A: Sample network target response intensity by trial type. Each horizontal line of the plot summarizes the network response in one target trial in terms of spikes per integration time step. Target trials are taken from the A-as-std (400), A-in-con (100), and A-as-dev (100) sequences, and sorted by sequence time (earliest trials are at the bottom of each section). B: Response summary, showing average network-wide spike counts in target trials in each sequence. The underlying data are the same as in panel A. Note how the A-as-dev response is very large, followed by an intermediate A-in-con response and a low A-as-std response. C: Distribution of single neuron responses (in mean spikes per trial) in all sequences. The 10 stimulated neurons responded more strongly and are omitted. The inset bars represent quartiles (thick bars) and first and last deciles (thin bars), and the median is indicated with a white dot. D: Distribution of single neuron response differences between A-as-dev and A-as-std, A-in-con sequences (all neurons, grey, no outline, no quantiles). The distribution shown as a black outline, which is also represented with quantiles as in panel C, excludes neurons that fire less than 0.1 spikes per trial in any sequence. E: Mean response sizes and , normalized by , across data sets (n = 120). In this and later boxplots, the median is indicated with an orange line, notches indicate the 95% confidence interval for the median calculated by bootstrap with 10000 iterations, boxes indicate the inter-quartile range (IQR), whiskers extend to 1.5 times the IQR or to the data extrema, whichever is less, and fliers indicate individual data points beyond the whisker end points. F: Deviance detection indices (DDI, see Eq (12)) across data sets.