| Abstract No.: | C-D3144 |
| Country: | Canada |
| | |
| Title: | INFLUENCE OF PREVIOUS CHOICES AND REWARDS ON SUPERIOR COLLICULUS ACTIVITY IS MODULATED BY BEHAVIORAL CONTEXT |
| | |
| Authors/Affiliations: | 1 Dhushan Thevarajah*; 1 Michael Dorris;
1 Queen's University, Kingston, ON, Canada
|
| | |
| Content: | Reinforcement learning uses the prior history of choices and rewards to adjust subsequent behavioral strategies. However, under some conditions, such as when playing strategic games, following a strict reinforcement learning process is counterproductive because it leads to predictable behaviors that can be exploited by a competitive opponent. This raises the question of whether reinforcement learning processes are malleable across behavioral contexts. To address this, we recorded the activity of saccade related neurons in the intermediate and deep layers of the superior colliculus, putatively involved in selecting/preparing upcoming choices, while awake behaving monkeys performed a force-choice and strategic free-choice task. During the forced-choice task a target appeared either within or opposite a neuron’s response field and successful acquisition of this target with a saccade resulted in delivery of a liquid reward 50% of the time. This allowed preparatory activity and subsequent saccadic reaction time on the current trial to be sorted based on previous choices and rewards. Despite the unpredictability of both choices and rewards on each trial, preparatory activity and upcoming saccadic reaction times were influenced by the recent history of each. To examine whether this activation consistent with reinforcement learning was under flexible control, we recorded from the same neurons while monkeys competed against a dynamic computer opponent in the competitive mixed strategy game ‘odds-evens’. Monkeys were free to saccade towards either of two targets presented on each trial but received a liquid reward only if they chose the same target as the computer opponent. Because predictability was exploited, the optimal strategy was to allocate choices stochastically independent of previous choices and rewards. Although the pattern of choices and rewards were similar in both tasks, the influence of each on superior colliculus preparatory activity was significantly attenuated during the strategy free-choice task. Our results suggest that motor preparatory activity, specifically, and reinforcement learning mechanisms, in general, are shaped by both previous choices and rewards. Differences in the degree to which these past events exert their influence in our two tasks show that the reinforcement learning process is not entirely automatic but can be flexibly tailored to meet context demands. |
| | |
| Back |
|
