Circuit mechanisms of self-organized cognitive strategies

项目摘要 数十年的心理学研究表明，工作记忆是有限的，人类只能容纳几个 同时在脑海中记着物品。然而，像计划和解决问题这样的认知任务需要访问 一次获取多条信息。为了克服这一限制，我们采用了助记策略，例如 将信息分成块，就像我们通常为了记住电话或社会保障而做的那样 数字。助记组块使我们能够灵活地组织在线信息，为我们提供了一个基本的 高级认知能力的构建块。当损伤或功能障碍时，就会出现组块障碍 累及背外侧前额叶皮质(DlPFC)，例如精神分裂症患者，严重 损害整体认知功能。因此，确定大脑如何组织信息是必要的 逐步了解高级认知的机制，以及这些机制在疾病状态下是如何出错的。 一个关键的挑战是，组织信息的策略是自己生成的，并且在 实验室环境。这一提议的一个核心创新是用来识别 猕猴的自发记忆组块。这一点至关重要，因为动物模型允许我们 使用先进的神经生理学工具询问大脑功能。在这里，我们将使用高密度、多站点 记录和靶向神经调节以了解组块记忆的电路机制 信息。先前的理论研究表明，组块来自压缩的工作记忆 充当神经速记的表示法，以降级为代价节省处理资源 一些原始信息。DlPFC中的神经元编码工作记忆中的项目，它们的动态被塑造 通过与基底节的反复相互作用。因此，我们假设皮质纹状体的相互作用促进 作为组块基础的工作记忆的有效重组。为了测试这一点，我们将调查dlPFC- 当猴子在自我组织的工作记忆任务中自发地分块信息时，纹状体动力学。 我们将记录大量的单个神经元和局部场电位，并动态解码 工作记忆中的表征被用来评估助记代码和皮质纹状体的相互作用 更改项目是否被分块的时间。此外，外源性刺激将检验 纹状体回路在促进记忆组块的形成中的作用。总之，这些实验将决定 大脑如何建立记忆区块以优化工作记忆性能。这将会使我们明白 高级认知的一个基本特征，以及这些机制的功能障碍如何导致 思维和记忆障碍。最后，了解优化认知功能的机制 生物系统可能会推动创造性的进步，从而优化人工智能系统的性能。

Project Summary Decades of psychology research have shown that working memory is limited, and humans can only hold a few items in mind at the same time. However, cognitive tasks like planning and problem solving require access to many pieces of information at once. To overcome this constraint, we enlist mnemonic strategies, for instance grouping pieces of information into chunks, as we commonly do to remember telephone or social security numbers. Mnemonic chunking allows us to flexibly organize information on line, providing a fundamental building block for advanced cognitive abilities. Chunking impairments occur when damage or dysfunction involves the dorsolateral prefrontal cortex (dlPFC), for instance in patients with schizophrenia, and severely compromises overall cognitive function. Thus, determining how the brain organizes information is a necessary step toward understanding the mechanisms of advanced cognition, and how these go awry in disease states. A key challenge is that strategies for organizing information are self-generated and highly variable in a laboratory setting. A central innovation of this proposal is the novel computational approach used to identify spontaneous mnemonic chunking in macaque monkeys. This is critical because animal models allow us to interrogate brain function with advanced neurophysiological tools. Here, we will use high-density, multi-site recording and targeted neuromodulation to understand the circuit mechanisms that chunk mnemonic information. Previous theoretical work suggests that chunks arise from compressed working memory representations that act as neural shorthand, economizing on processing resources at the cost of degrading some original information. Neurons in dlPFC encode items in working memory, and their dynamics are shaped by recurrent interactions with the basal ganglia. Thus, we hypothesize that corticostriatal interactions promote the efficient reorganization of working memory that underlies chunking. To test this we will investigate dlPFC- striatal dynamics when monkeys spontaneously chunk information in a self-organized working memory task. We will record large numbers of single neurons and local field potentials, and dynamically decode representations held in working memory to assess how mnemonic codes and corticostriatal interactions change when items are or are not chunked. In addition, exogenous stimulation will test the causal role of striatal circuits in promoting the formation of mnemonic chunks. Together, these experiments will determine how the brain establishes mnemonic chunks to optimize working memory performance. This will shed light on a fundamental feature of advanced cognition, and how dysfunction in these mechanisms could give rise to disorders of thought and memory. Finally, understanding mechanisms that optimize cognitive function in a biological system may fuel creative advances that optimize performance in artificial intelligence systems.