Figure 2.

Single unitresponses

(A) Single unit responses under different task conditions, as indicated by different colors. Two shaded areas represent the period of contextual cue stimulus presentation and sensory stimulus presentation, respectively. Left panel: Responses of units showed pure selectivity to the task-related variable. Unit 1 preferred the color cue, unit 2 preferred the motion cue, unit 6 preferred green, unit 7 preferred red, unit 10 preferred the leftward direction, and unit 11 preferred the rightward direction. Right panel: Different task-related variables were mixed. Unit 3 showed mixed selectivity for the color cue and motion cue. Units 4 and unit 5 showed mixed selectivity for both the context cue and sensory stimulus. Unit 8 showed mixed selectivity for two task variables (green and red) for the color context. Unit 12 showed mixed selectivity for two task variables (left and right) for the motion context. Unit 9 showed mixed selectivity for three task features, such as the color context cue, green, and red sensory stimulus. Similarly, unit 13 showed mixed selectivity for the motion context cue, leftward direction, and rightward direction.

(B) The percentage of units in the trained RNN that were activated to perform a delay context-dependent decision-making task, averaging over all 16 trial conditions and 20 trained RNN models.

(C) Among all the activated units, the percentage of units that was selective to the context cue, sensory input, and their combinations in color and motion context, respectively. Error bar indicates SEM over 20 different training configurations.