Conference on Cognitive Computational Neuroscience

Relational memory connects people, events, and objects through spatial and temporal relationships, enabling us to navigate the past and plan for the future. This study investigates how we efficiently form relational representations in our memory with a sequential memory task, where participants predicted the next movement of symbols based on rewards. A computational model with reinforcement learning revealed that participants focused on relevant information, reduced the dimension of the representational space, and inferred latent causes through contextual knowledge under the constraints of the limited capacity of working and short-term memory. This study provides insights into how we turn daily randomness into a logically connected life.