Information Flow (Transfer Entropy) between Sensory Input, Neural Representation of Saliency, and Motor Variables in <i>Roboto</i>

<div><p>(A1) Transfer entropy between array <i>I_R</i> (variable S) and pan-tilt amplitude (variable M). Series of plots show maps of transfer entropy from S to M (S → M) and from M to S (M → S) over visual space (55 × 77 pixels), calculated for offsets between −7 (“M leading S”) and +7 (“S leading M”) time steps. Plots show data for conditions “fov” and “rnd.” The gray scale ranges from 0.0 to 0.5 bits (for all plots in panels A1 and B1).</p><p>(A2) Curves show transfer entropy for five individual runs (thin lines) as well as the average over five runs (thick lines) between the single central pixel of array <i>I_R</i> (S) and pan-tilt amplitude (M), for directions M → S (black) and S → M (gray).</p><p>(A3) z-Score maps of significant image regions (plotted between z = 0 and z = 6). The z-scores are expressed as number of standard deviations above background at time offset +1 (S → M) and −1 (M → S). Mean and standard deviation of background is calculated from transfer entropy values at maximal time delays (−7,+7 time steps).</p><p>(B) All three panels have the same format as (A), but the neural activations of the saliency map <i>Sal</i> are substituted as variable S (11 × 11 neural units).</p>