1. The Dynamics of Feedback-based Learning is Modulated by Working Memory Capacity
- Author
-
Markus Ullsperger, Humann J, and Adrian G. Fischer
- Subjects
Computer science ,Working memory ,Human–computer interaction ,Dynamics (music) - Abstract
Research suggests that working memory (WM) has an important role in instrumental learning in changeable environments when reinforcement histories of multiple options must be tracked. Working memory capacity (WMC) not only reflects the ability to maintain items, but also to update and shield items against interference in a context-dependent manner; functions conceivably also essential to instrumental learning. To address the relationship of WMC and instrumental learning, we studied choice behavior and EEG of participants performing a probabilistic reversal learning task. Their separately measured WMC positively correlated with reversal learning performance. Computational modeling revealed that low-capacity participants modulated learning rates less dynamically around value reversals. Their choices were more stochastic and less guided by learnt values, resulting in less stable performance and higher susceptibility to misleading probabilistic feedback. Single-trial model-based EEG analysis revealed that prediction errors and learning rates were less strongly represented in cortical activity of low-capacity participants, while the centroparietal positivity, a general correlate of adaptation, was independent of WMC. In conclusion, cognitive functions tackled by WMC tasks are also necessary in instrumental learning. We suggest that noisier representations render items held in WM as well as tracked values in instrumental learning less stable and more susceptible to distractors.
- Published
- 2020