Figure 4

EMD model with added spatial filter and threshold function reproduces the experimental results.(a) The structure of our model. First, the visual stimulus is filtered with a 2D Gaussian function ( = 7.5°) and the filtered signal is passed through a sigmoid function (threshold function). The sigmoidal output is processed by the 2D array of a 2-Quadrant-Detector model (see Methods). HP, temporal first-order high-pass filter (τ = 250 ms); DC, passing 10% of the original signal; LP, temporal first-order low-pass filter ( = 150 ms); M, multiplication; Sigma (Σ), nonlinear integration. (b) Putative 2D appearance of the LED display from the viewpoint of a tethered fly (top). The spatial distribution of dendritic integration for the model cells (bottom). We constructed the vector field of spatial weight factors adjusted for the receptive field of the right HS cell⁴⁴. The length and orientation of each vector indicates the level of sensitivity and the preferred direction of local motion detector, respectively. (c) Average responses of the model cell (10 trials) to visual stimuli with various SNR in both PD and ND stimulations at a temporal frequency of 1 Hz (black). Average membrane potentials of 10 cells are also presented (gray). The gray-shaded region indicates when visual stimuli were in motion. (d) Mean AUC (red, contains both the spatial filter and threshold function as illustrated in a; gray, contains either the spatial filter or threshold function or none of the two). (e) Mean normalized MRD (10 trials) for each model (blue, contains both the spatial filter and threshold function [filled circles] and HS cells illustrated in Fig. 3i [open circles]; gray, contains either the spatial filter or threshold function or none of the two).