Fig. 4: Results for generation across all pairs of brain regions.

a In this evaluation, training data and prediction, generation, data are also prepared from different regions. b In the case of a, training data (first 17 minutes) is obtained from region A and test data (second 17 minutes) from region B. c This panel shows the result of predicting the firing rate in a particular case, where the x axis shows the firing rate in the original test data and the y axis shows the firing rate in the generated test data by training Multilayer LSTM. d This panel shows the result of predicting the synchronization score in a particular case. Again, the x axis is the synchronization score in the original test data, and the y axis is the Synchronization score in the generated test data. This two-dimensional distribution was expressed in terms of r-θ rotational coordinates, and panel. e depicted the density distribution of the number of data in 0-π/2 with respect to θ, i.e., in the first quadrant. Finally, panel f is the density distribution of the number of data in π-3/2π with respect to θ, or the third quadrant. In particular, we have confirmed that the distribution is restricted to the third quadrant when the output is from inhibitory cells. The sharpness of the peaks in these histograms (e, f) was evaluated by sharpness. g is the correlation between generated and truth values in firing rates for all cells, h for inhibitory cells, and i for excitatory cells. x axis is the region index of the original data and y axis is the region index of the predicted data. The correlations of firing rates in all those pairs are plotted as color maps. In addition, these color maps are sorted based on hierarchical clustering. j, k are the color maps at sharpness in the first and third quadrants, respectively. The point that the sorting is based on hierarchical clustering is the same as in the case of the color map of firing rates.