Fig. 5: The neuron population predicts the crows’ impending number of pecks.

a Accuracy of within- and across-protocol classifier prediction performance. Average accuracy for prediction of target number from firing rate of the sampled NCL population during the motor planning period. Within- (trained on trials from a protocol and tested on new trials from the same protocol) and across-protocol performances (trained on trials from a protocol and tested on new trials from the other protocol) are shown. Dotted lines represent 95th percentile of shuffled label classifier performance. Error bars indicating SEM are too small to be depicted. b Performance of SVM classifier. Curves show the classification performance for each target number (color coded). Error bars indicating SEM are too small to be displayed. c Confusion matrix showing the accuracy of the classifier predicting the number of impending pecks in relation to the instructed number of pecks. The main diagonal describes correct classifications. Values from rows of the confusion matrix result in the curves in b. d Classifier accuracies for predicting correct and incorrect number of pecks when the classifier was trained on firing rates of correct trials from the motor planning period. Dotted lines represent 95th percentile of shuffled label classifier performance. Error bars indicating SEM are too small to be depicted. e Classifier accuracies (spanned in three-dimensional space) for predicting whether the crows would generate less (−1 error) or more pecks (+1 error) than instructed (correct) based on firing rates during the motor planning period. The correct number of pecks and both error types are predicted above chance (33% for three classes; indicated by dotted line). Source data are provided as a Source Data file.