Fig. 8: Behavioral error history is signaled by value neurons during reversal learning prior to performance increase.
From: Dynamic perceptual feature selectivity in primary somatosensory cortex upon reversal learning
a Responses of a value neuron in hit trials during reversal learning but before the behavioral performance increases. Hit trials preceded by a false alarm trial (FA, light green) have larger responses than hit trials that were preceded by another hit trial (dark green). b Average performance curves across mice realigned to the expert criterion (70% correct, dotted line) post-reversal, calculated with a rolling window of 200 trials (N = 10 mice containing value neurons; Supplementary Table 1). c The corresponding average hit and FA rate averages. d The corresponding average calcium signals (ΔF/F0) in hit trials that followed upon another hit trial (dark green) or upon a FA trial (light green) for the various neuronal classes: non-selective (N = 412, 12 mice), value (N = 35, 10 mice), choice (N = 24, 6 mice), texture selective (N = 33, 7 mice), gained (N = 144, 12 mice), and lost selectivity (N = 152, 12 mice). The arrowheads in c and d indicate the trials at which the 95% confidence intervals separate for hit and FA rates (black), and for the responses in the hit–hit vs. the FA-hit trials (red). e Calculated mean error history representing the normalized difference between the two types of hit responses (Methods section). f Percentage of neurons per class with significant error history rates (white line) above the cumulative distributions of error history values. Shaded areas in b–e represent 95% confidence intervals.
