Figure 6. A simple network model generates signatures of speed pressure via global gain modulation alone.

(a) The model represented as a simple two-layer network in which choice-determining accumulation units are subject to recurrent excitation (orange), lateral inhibition (red) and leakage, and all connection strengths are modulated by global network gain (green). (b) The non-linear transfer function relating a unit's input to its corresponding output. The global gain parameter g determines the slope of this function. (c) g as a function of elapsed decision time in each speed regime. Informed by the pupil diameter results, there is a static offset in g at the beginning of a DL trial coupled with a time-dependent increase in the DL but not FR regime. All other model parameters are fixed across regimes. Main plot shows g time courses when the effective time constant of accumulation τ is unconstrained and takes a long value (555 ms); inset shows evolution of g when τ is constrained to equal 167 ms (see ref. 52). (d,e) Observed and fitted RT distributions and conditional accuracy functions. Points in e depict mean±s.e.m. accuracy of trials sorted by RT into 25 equal-sized bins. (f,g) Activation time-courses of the winning and losing accumulator units, simulated using fitted model parameters with unconstrained τ and aligned to motion onset (f) and decision commitment (g, sorted by RT into 3 bins). Global gain modulation qualitatively produces both a pre-motion offset in activation of both accumulators and a time-dependent increase in common activation of both accumulators under deadline. (h) Activation of the losing accumulator at decision commitment plotted as a function of commitment time for each speed regime.