Figure 2. Our model accounts for the behavioral and neural findings in Bromberg-Martin and Hikosaka (2011).
(A) The task in (Bromberg-Martin and Hikosaka, 2011). On each trial, monkeys viewed a fixation point, used a saccadic eye movement to choose a colored visual target, viewed a visual cue, and received a big or small water reward. The three potential targets led to informative cues with 100%, 50% or 0% probability. (Bromberg-Martin and Hikosaka, 2011; reproduced with permission) (B) Monkeys strongly preferred to choose the target that led to a higher probability of viewing informative cues (Bromberg-Martin and Hikosaka, 2011; reproduced with permission). (C) The activity of lateral habenula neurons at the predicting cues following the 100% target (predictable) were different from the case where the cues followed the 50% target (unpredicted) (Bromberg-Martin and Hikosaka, 2011; reproduced with permission). The mean difference in firing rate between unpredicted and predictable cues are shown in case of small-reward and big-reward (the error bars indicate SEM.). (D) Our model predicts the preference for more informative targets. (E,F) Our model’s RPE, which includes the anticipation of rewards, can account for the neural activity. Note the activity of the lateral habenula neurons is negatively correlated with RPE.
Figure 2—figure supplement 1. Our model can capture the preference of info targets with a wide range of parameters.
Figure 2—figure supplement 2. RPE-boosting of anticipation is necessary to capture the choice preference of monkeys reported in Bromberg-Martin and Hikosaka (2011).
