Fig 2. A mechanistic model of MGS performance.

a) MGS performance is modeled as the transfer over time of a topographic WM representation into a topographic representation of an oculomotor response. In addition to accumulating Gaussian random noise, key sources of neural variability include two independent gain modulating signals, one that affects the amplitude of WM activity, and a second that modulates the response threshold of the oculomotor population. b) Depicts the state of simulated neural activity within WM and oculomotor populations at an arbitrary time point during a trial. The activity of each neuron codes either for the location of a remembered target or a saccade. The height of each bar represents the activity of a single neuron. The neuron representing the location closest to the remembered target is rendered in red. Gain variability affects WM by multiplicatively scaling the population activity (top). The effect of gain variability on the oculomotor population is the modulation of the response threshold (middle). The peak of activity within WM and oculomotor populations coincides with the remembered location of the stimulus and subsequent memory-guided saccade (bottom). c) An example time course of accumulating activity within the oculomotor population for a trial. Each line corresponds to the state of a particular oculomotor neuron across time; coloring as in b. A simulated memory-guided saccade is performed once any neuron in the oculomotor population passes the response threshold.