Fig. 1. Stimuli, data analysis and modeling. (A) Stimuli consisted of Gabor patches that drifted in random directions at different locations within the receptive field of the neuron. During preliminary testing we presented 4x4 patches that covered the entire monitor (low-contrast patches in left-most frame) to obtain a vector map for a 80°x80° portion of the receptive field (see Fig. 2B for an example). Subsequent testing only involved two patches (high-contrast), independently changing direction every 220 ms (in this figure the low-contrast patches indicate the other possible positions, but were not presented in the two-patch stimulus; in the full version of the stimulus (showing all patches) they were presented at high contrast). (B) We computed response surfaces for all possible directions of the patch at position 1 and time corresponding to when the response was measured (broken blue circle in A), together with all the possible directions of the preceding patch at position 2 (broken red circle in A). The specific combination shown in A is indicated by ^* in B (note that this panel does not show real data). A similar analysis was carried out for the preceding patch at the same location (broken white circle in A). (C–E) Schematic descriptions of three models that were tested in this paper. Proceeding from left to right, the nonlinear transducer (red) is placed at progressively earlier stages within the model. See Materials and methods for details of model implementation.