The Nature of Working Memory Representations

点击翻译按钮获取中文摘要

PROJECT SUMMARY Because most high-level cognition depends on working memory (WM) and its dysfunction causes a host of cognitive impairments, researchers have spent decades trying to understand the neural mechanisms that support WM. Recently, using sophisticated computational neuroimaging approaches researchers have repeatedly decoded the contents of WM from patterns of neural activity in a widely distributed number of brain areas. The format of the WM representation in early visual cortex, for instance, might have the same “sensory- like” properties as the visual stimulus. Although sensory-like representations are less likely in higher-order brain areas, the nature of these alternative representations has yet remained impenetrable. This gap in our knowledge is a critical problem because a host of psychiatric and neurologic disorders stems from a primary WM dysfunction. Our long-term goal is to understand the mechanisms by which neural populations across the brain encode WM representations, and how we might develop strategies to mitigate WM problems that impact the quality of cognition. Our overall hypothesis is that what one sees and what one stores in WM can be distinct and that distinction differs across the cortical hierarchy. The central aim of the project is to develop incisive data analytic approaches that will reveal the nature of what is actually being encoded in the neural population dynamics underlying WM storage. The rationale for the proposed research is that as we better understand the neural mechanisms of WM, a strong theoretical framework will emerge within which strategies for understanding and treating cognitive dysfunction will emerge. We test our central hypothesis by pursuing two specific aims. 1) We will model the neural population dynamics that code for distinct formats of WM representations. 2) We will identify when and where neural populations encode WM representations that are abstractions of sensory features. Strong preliminary data demonstrate the feasibility of proposed work as well as initial support for the hypotheses. Under Aim 1, using novel dimensionality reduction techniques suggest that neural populations code for both a representation of the memorized stimulus and a representation of the specific stimulus feature that is task relevant. Under Aim 2, using novel means to model and visualize WM representations revealed that neural populations that are traditionally thought to encode visual stimulus features in WM also store abstractions that can bear little resemblance to the original visual stimulus. Overall, the proposed work will generate the data needed to unmask the representational format of WM across the cortical hierarchy. The approach is innovative because it uses direct and unbiased computational approaches to model and visualize the representational format of WM in ways that have not been applied in neuroimaging. The proposed research is significant because it will provide key insights into the nature of WM representations in the human brain, in addition to providing new targets for cognitive remediation in psychiatric, neurologic, and geriatric populations.