Poster Session A, Saturday, March 23, 1:30 pm - 3:30 pm, Pacific Concourse
Development of cortical and sub-cortical components of learning: A computational analysis
Maria Eckstein1, Sarah Master1, Ronald Dahl1, Linda Wilbrecht1, Anne Collins1; 1University of California, Berkeley
Different brain systems mature at different rates, but little is known about the maturation of cognitive capacities that depend upon interactions across these systems. We sought to address how reinforcement learning processes and cognitive control, which rely on early-maturing subcortical regions and later-maturing frontal regions, respectively, develop during adolescence. We tested 197 youths (91 male; ages 8-17) and 53 adults (ages 25-30) in a probabilistic reversal learning task. Participants used noisy, binary feedback to learn which of two choices was most likely to lead to reward. The best option changed occasionally and unpredictably, engaging cognitive control and reinforcement learning. Children (ages 8-12) switched strategies more quickly than adolescents and adults after negative feedback, and out of all age groups, teenagers (13-18) employed the strategy closest to Bayes optimal: they were least likely to impulsively change strategies in the face of spurious absence of reward. Computational modeling employing Bayesian inference and reinforcement learning suggests that the age-related differences in behavior stem from decreasing decision noise with age, increased sensitivity to negative feedback in children compared to the other age groups, and better estimation of the underlying task contingencies by adolescents and adults. This research highlights changes in cognitive abilities associated with both subcortical regions (reward sensitivity) and frontal regions (estimation of task contingencies), suggesting a complex developmental trajectory and interplay of different systems.
Topic Area: THINKING: Decision making