It is currently believed that remembered visual target locations are stored in eye frame and updated across eye movements. Neurons in brain areas associated with saccade generation have shown transient receptive field remapping prior to and during saccades. The question remains as to whether these receptive fields spread or jump (Wurtz & Sommer 2005).
We used a simple 3-layer neural network with recurrent connections between units in the hidden layer to examine the temporal dynamics of updating using the full 3-D geometry of eye rotations. We found that the network was able to perform the required updating task, and that it did so by remapping receptive fields of both output and hidden layer units. While the network was trained to generate a hill of activation in the output layer before and after the saccade, no constraint on the behavior of the network during the saccade was made. The network developed a moving hill of activation in the output layer during the saccade, but with suppressed activation magnitudes.
This suppressed moving hill reconciles previously conflicting findings of moving and jumping hills. The mechanisms observed appear to be a viable model for how trans-saccadic spatial updating is done in the brain.