人类的视觉系统可以同时接收外界环境中的多种视觉信息,并以记忆表征的形式暂时存储于工作记忆之中,以便指导后续的认知活动.但大脑只能维持有限数量的记忆表征,表现出工作记忆的容量有限性.同时,我们对视觉刺激的内部表征是充满噪音的,即对记忆表征的维持存在精确度的变化.本文从大脑神经活动的神经振荡特性角度出发,系统阐述了人类视觉工作记忆的表征维持阶段、记忆容量和记忆精确度的神经振荡特征,以及相关的理论建模,并对该研究领域的发展做了展望.

Human visual system receives many cluttered stimuli simultaneously, but only a few of visual representations can be held in the mind to guide subsequent goal-oriented behaviors, which indicates the limited capacity of visual working memory. Meanwhile, the neural representations of visual stimuli can be noisy and hence the memory precision is varying. This review systematically illustrates the roles of neural oscillations in maintaining visual working memory and neural mechanisms underlying the capacity and precision. In the end, several potential directions of this field for future studies are discussed. A large amount of researches demonstrates that during the maintenance of visual working memory representations, the theta band oscillations can facilitate the neural information communications between brain regions throughout a relative long distance, in order to achieve the top-down cognitive control; the alpha band oscillations, which was initially considered as an electrophysiological correlate of cortical idling, are now generally treated to reflect functional inhibition or non-engagement of a given brain regions, in order to gate neural information to the task-relevant regions; as for gamma band, the high frequency oscillation property make it suitable for the integration of multiple object features in the local neural network and hold the relevant representations. Therefore, theta and gamma band engage in the maintenance of working memory representations directly, while alpha oscillations serve in an indirect way of functional inhibition. In addition, the cross frequency coupling and neural synchronization enable the collaboration of different brain regions in the working memory network. The contralateral delay activity（CDA） is widely acknowledged to be an online electrophysiological marker of working memory capacity. In addition, the alpha oscillations in irrelevant regions also have a strong correlation with individual capacity. Early studies that aimed to explore memory capacity limit introduce the slot model of working memory, which regards working memory as few fixed discrete slots and cannot be divided between items. However, more and more experiments that are designed to study memory precision come to conceptualize working memory as a limited resource that can be shared between different representations, and put forward the resource model of working memory. One of the major differences of these two models is the change of memory precision with memory load increasing, so the study of memory precision is conductive to figure out the working memory model. Although several behavioral studies have demonstrated that both capacity and precision are two distinct aspects of working memory, some electrophysiological researches propose that CDA and alpha band oscillations can also represent the memory precision. Therefore, more studies would be needed to explore the neural substrates of memory precision. In addition to the passive maintenance of working memory representations, the active manipulation of memories, such as updating, is a potential direction of working memory studies, since it can give us a whole picture of how the memory resources are dynamically distributed among objects. Moreover, neural decoding of objects is also an important research perspective, which can directly manifest the storage of memory representations.