Re-evaluating Models of Visual Working Memory in Light of Interference Effects

Abstract

The cognitive processes responsible for temporarily maintaining and manipulating visual information, collectively referred to as visual working memory (VWM), are essential for understanding our environment and acting on it in a flexible and goal-directed manner. A substantial body of research indicates that stored stimulus representations in VWM are susceptible to interference from subsequent sensory input, and conversely, that ongoing VWM storage can interfere with the perception and processing of new visual input. However, there remains no clear consensus as to how these interactions manifest. This is partly because the interpretation of observed interference effects depends on the level of analysis and on which theoretical model of VWM storage is presupposed. The present thesis reviews findings on VWM interference in light of each other and in light of the currently dominant neurocognitive models of VWM storage in an attempt to evaluate their mutual compatibility, identify potential points of convergence or conflict, and explore ways in which these models might be integrated or revised to better account for the observed data. A number of theories are addressed regarding the nature of VWM storage, it’s capacity, and it’s susceptibility and robustness to interference. Ultimately, the thesis proposes an integrated multi-level framework of VWM storage. In this framework, VWM simultaneously is proposed to store a broad, low-fidelity gist representation that tracks overall properties of the visual scene, and a discretely limited number of distinct, high-fidelity item representations. Each item representation is proposed to be manifest as a synaptic enhancement of the responsiveness of the sensory neurons tuned to the visual features of that item. The consolidation and reactivation of these latent item representations are proposed to be orchestrated by a dynamic oscillatory system.