Figure 3.

Pointing errors as a result of saccades using the network in Figure 1B. We are leaving out the recurrent connection in the Hidden layer at this point, so latencies are still hard-coded. FEF output is combined with eye position information in the EP layer to create a BC layer representation of the object that guides the arm. Conventions as in Figure 2. In one model (blue trace), the EP signal was updated at known proprioceptive latencies, 50 ms after the saccade. In a second model, the EP signal was updated in synchrony with the arrival of CD information at the FEF layer (orange trace). Both networks outperformed the models in Figure 2, due to the addition of the BC layer. And when the latency of the EP update matched the latency of the CD signal, pointing was quite accurate through the course of the trial, nearly matching ideal performance (dashed line).