Fig. 8:
Lower-dimensional representation of activation data (no reservoir) and classification accuracy. Accuracy is shown as a function of the number of top plus bottom components. The magenta line indicates the highest performance using the full-dimensional activation data. For working memory (A), classification accuracy with the lower-dimensional representation reaches 95% of the full data with 6 dimensions. However, for the theory of mind (B), 28 dimensions (as opposed to 12 when using reservoir data) of the lower-dimensional representation. Error bars show the standard error of the mean across validation folds.