Visual working memory load induces similar representational structures in a distributed cortical network

摘要

Visual working memory load induces similar representational structures in a distributed cortical network Working memory is a cognitive process for temporarily holding information, which is stored in a distributed neural system by neu- ral population coding, to guide task-relevant actions. However, the effects of working memory load on the coding characteristics in the neural system still remain unclear. Here, we investigated the cod- ing characteristics under different working memory loads in the distributed neural system using neural representational structures of visual stimulus. We used n-back working memory task fMRI data from Human Connectome Project dataset. We specified the neural system participating in the visual object working memory process by using the general linear model analysis. We constructed the neural representational structure using inter-stimulus neural rep- resentational distance matrix (IS-RDM), which is calculated as one minus Pearson correlations between multi-voxel fMRI activity pat- terns of all stimulus pairs in each region. We made IS-RDM of probe objects, which are externally presented, for both task conditions and an IS-RDM of target objects, which are internally maintained, for the 2-back condition. We calculated the mean and standard deviation of IS-RDM to estimate the size and shape of the struc- ture. We compared them between 0- and 2-back conditions in the distributed neural system. We found that the mean and standard deviation of IS-RDM are different between the 0- and 2-back con- ditions in most regions. Next, we found that Inter-regional IS-RDM similarity in the distributed neural system is significantly increased in the 2-back conditions for both target and probe objects. Finally, we found that the Inter-regional IS-RDM similarity is significantly increased in correct trials than in false trials for target objects of 0-back and probe objects of 2-back conditions. These results indi- cate that the working memory load modulates the size and shape of the neural representational structures in the distributed neural system to shape a similar structure, which induces better working memory.