Process-switch in Working Memory Impose Significant Cost to Its Performance

Abstract

Working memory is widely accepted to be limited in capacity, but how various aspects of memorized items affect working memory capacity and performance continues to be of great interest to many neuroscientists. A recent breakthrough in the field shows that working memory resources are flexibly distributed among numbers of items that need to be maintained. The allocation of resources is modulated by how much attention the item is given. Although the new framework presents useful insights, a more natural setting in working memory utilization often requires more than simple maintenance of multiple items—it requires updates that involve manipulation of memory. This thesis contains a series of investigations of memory updates on working memory performance using the established techniques of psychophysics and mixed linear regression. First, we perform a working memory update task on 102 individuals. We find a unique cost previously underestimated in the working memory behavioral paradigm—the cost of switching between working memory process modes. We show that the switch has a significant and substantial effect on the error, and it has a positive multiplicative relationship with the number of other concurrently maintained representations. Second, we show that the standard deviation estimate ( ) from a classic decision-making task is correlated with the estimate from the orientation estimation, supporting the use of SD estimate from the decision-making task in place of one from estimation. Additionally, we show that the decision-making paradigm allows a trial-to-trial measure of performance in response time, which has important implications for future analysis. Third, we develop a unique experiment paradigm inspired by the given experiments and test its potential for uncovering a novel and essential aspect of the switch between different operational modes of working memory (maintenance and manipulation). We find some indication that the source of the main cost in performance is the process-switch itself. Simultaneously, the processes—retrieval and manipulation of an item—do not affect working memory performance as expected. We show that 1. the cost of manipulating a target is increased when the non-target item is retrieved, 2. the target items memory is impaired when the non-target item is manipulated, 3. the cost of manipulating an additional item is negligible if the other item in memory is already manipulated. The results suggest that modulating between the maintenance network'—activated by the retrieval process—and the manipulation network'—activated by the manipulation process—incurs the greatest cost in working memory performance. In conclusion, our findings suggest that the process-switch and not the processes themselves incur a great cost in working memory. More importantly, we devised an exciting novel experimental paradigm that can be utilized to study the top-down executive control of modular working memory more readily.