(A) model study on the neural activity patterns with the variation of visual working memory load

Abstract

Visual working memory provides temporary active retention of visual information for supporting our behaviors. It is reported that human working memory performance degrades as the memory load increases. This has been studied in relation to memory capacity. The model known as the slot model considers that working memory is limited by the number of ‘slots’, which can remember items. On the other hand, the resource model considers working memory to be a limited resource that is distributed flexibly between all items to remember. However, the two models compete in how they explain working memory. Here, we study how neural activity patterns in the visual pathway vary with the visual memory load during a working memory task. More specifically, we undertake a computational simulation of the five-layered hierarchical structure of the visual pathway from the retina to the frontal cortex. We also examine human memory performance for comparison with the simulation results.

We modeled the five-layered hierarchical structure of the visual pathway, which has neural activity patterns within the retina and regions of the occipital and parietal lobes and the frontal cortex. We assumed that two parameters modulate the model, i.e., the degree of convergence and the nonlinear response function of neural activity. Each layer simulates the modulation of these parameters to transmit visual information to next layer. This modulation enables the model to describe the previous fMRI reconstruction across the brain regions under memory loads one and two.

We examined the usefulness of the model under other stimuli that were newly designed to have different memory loads. Initially, we tested human memory performance with the newly designed stimuli. Subjects were asked to memorize a first image and then report whether a second image was identical or different from the first. We found a correlation between human performance and the simulated neural activity patterns. Our results also suggest that this neural activity pattern is correlated with the memory load regardless of the type of stimulus.

Our visual working memory model shows that neural activity patterns are correlated with different memory loads with modulation of the feedforward convergence and cell response function. The quantitative match between neural activity patterns and an increase in the memory load supported the resource model of working memory. Furthermore, we could predict human memory performance and neural activities within brain regions according to the memory load under certain stimuli using our model.