Scatter plots of the responses to noise and grating stimuli of three populations of broad-spiking cells, (A) before recovery and after recovery during locomotion viewing (B) noise or (C) gratings. Data from noise+run mice shown in B, D, E, H; data from grating+run mice shown in C, F, G, I. We used the earth mover distance algorithm (Y Rubner et al., A metric for distributions with applications to image databases. Proc. IEEE International Conference on Computer Vision, Bombay, India, January 1998, pp. 59–66) to calculate the fictive correspondence between the 500 most responsive individual neurons of A and either B or C as the minimum total distance required to move the points of one distribution into the other, cartooned for simple cases in D for noise and G for grating responses. The flow field of movement vectors (arrows in D and G) is plotted for each recovery condition as a heat map of the summed components along the noise response axis (E and F) or grating response axis (H and I). Color bar inset in H indicates four spikes/sec along noise F1 response axis in E and F and 15 spikes/sec along grating response axis in H and I. These heat maps show that the effect of noise stimulation was to increase the response to noise stimuli selectively in neurons that were initially more responsive to noise than to gratings, as indicated by warm spots above the diagonal in E and lack of warm spots in H. The effect of grating stimulation was to increase the response to grating stimuli selectively in neurons that were initially more responsive to gratings than to noise, as indicated by warm spots below the diagonal in I, and paucity of warm spots in F, (p<0.001 for all comparisons E vs H, F vs I, Chi-square). Computer source code used for analysis and source data are supplied in supplement.