Neural correlates of reinforcement learning, working memory and choice dynamics: an EEG study on the RLWM task.

Recent research has aimed both to disentangle the cognitive mechanisms that support learning—particularly the roles of reinforcement learning (RL) and working memory (WM)—and to characterize the dynamic processes that govern decision-making during learning. These two endeavors were integrated with the reinforcement learning working memory Linear Ballistic Accumulator (RLWM-LBA) model. The RLWM-LBA offers a unified framework to investigate how RL and WM contribute to learning and how these processes unfold over time to shape choice behavior. Specifically, it remains unclear whether neural processes of evidence accumulation identified in perceptual decision-making are also present in learning and how such a process is associated with RL and WM. To address these questions, we analyzed electroencephalography (EEG) data from 510 participants performing a stimulus-response association task. Using a Bayesian hierarchical RLWM-LBA model combined with event-related potential (ERP) analyses, we explored whether neural signatures of RL and WM persist when captured with RLWM-LBA model and whether a neural signal of evidence accumulation, the centro-parietal positivity (CPP), could be identified within this learning context.Our findings confirm previous reports by identifying distinct neural correlates for RL and WM. Importantly, we demonstrate a neural signature corresponding to internal uncertainty about learned action policies that closely resembles CPP signals documented in perceptual and preference-based decisions. These results provide robust evidence that the neural dynamics of decision-making during learning adhere to established evidence accumulation mechanisms and highlight how RL and WM processes dynamically shape learning and choice behavior.