Skip to main content

Advertisement

Figure 2 | Behavioral and Brain Functions

Figure 2

From: Evidence that the delay-period activity of dopamine neurons corresponds to reward uncertainty rather than backpropagating TD errors

Figure 2

When visual stimuli predict reward at p = 0.5, the delay-period activity of dopamine neurons increases with the discrepancy between potential reward magnitudes, but the phasic response to reward delivery is independent of reward magnitude. A. The onset of each of three distinct visual stimuli was followed by either of two potential liquid volumes with equal probability. Histograms represent the average activity of 35 dopamine neurons; these neurons were not selected for the presence of any task-related modulation. Delay-period activity increased with the discrepancy between potential liquid volumes. See figure 4 of Fiorillo et al [3] for a full summary of this data. B. Population histograms showing the average response of dopamine neurons to the delivery of reward in the same experiment as illustrated in A (n = 57). Data are from figure 4 of Tobler et al [8], and include data from 22 neurons tested with trace conditioning (as described in [3]) that were not included above in panel A. No differences were observed in the phasic activation to reward in trace versus delay conditioning. The critical observation here is that the delay-period activity varies (from top to bottom) in panel A, but the phasic 'prediction error' response to reward does not (as shown in B). This data is inconsistent with the proposal of Niv et al., according to which the delay-period activity shown in each panel in A should scale with the corresponding phasic prediction error response shown in B. Since the two responses do not scale together, it appears that the delay-period activity cannot be accounted for by the backpropagation of prediction errors.

Back to article page