Tracking the emergence of internal models

摘要 人类和动物认知的核心是内部模型的概念： 关于世界结构及其启示的知识，使预测和 规划这些模型的存在是经验的基础。引导我们在 在这个世界上，我们接收到的原始即时感官信息是高度贫乏的， 动态相对于丰富的，有组织的，稳定的和详细的经验性质。学习和 也许部分先天的先验允许维持对世界的真实再现 快速选择和整合与任务相关的重要信息。在这 建议，我们的目标是通过记录多个世界模型来探索世界模型的神经实现 灵长类动物在自然环境中导航时的大脑区域。通过建模， 在单个复杂序列期间跨多个实验室的协调记录的分析 任务，我们将制定一个全面的了解如何这样的模型发展，取决于积极的 与世界接触，影响感知。在我们的项目中，三只猕猴 神经生理学实验室，在不同的大脑区域具有独特的尖端专业知识， 将与具有不同专业知识的计算神经科学家合作， 追求重叠的目标。我们提出了一种新的合作策略，其中所有三只猕猴 神经生理学实验室将调查一个共同的导航任务，并记录从不同的，但 重叠的区域子集：下颞叶皮层（IT），运动皮层（MC）和前额叶 皮质（PFC）。我们首先关注IT-HC（Tsao）、MC-HC（Orsborne）和PFC-HC（布法罗）， 假设HC，凭借其解剖结构，其在情节记忆中的基本功能 形成，以及它在协调主观体验中的主导作用，构成了一个中心枢纽 来代表大脑内部的世界模型。为了帮助解释这些新数据，我们将 与分析合作开发和探索预测处理的网络模型 理论团队Fairhall、Mihalas、Rao和Shea-Brown，在以下方面具有互补的专业知识： 神经编码、预测编码和网络动力学将相互作用， 模型框架，并与每个实验室单独进行数据分析。主要 我们将解决的问题包括：这样一个“世界模型”是如何发展的，它是如何发展的， 在大脑中的表现？世界模型对人类的动态有什么影响？ 感觉处理和行为学习的结构知识如何与自我整合？ 在探索环境的过程中，？在记忆过程中这些信息是如何被提取的- 导航到一个目标？

Abstract Central to human and animal cognition is the idea of internal models: an internal repository of knowledge about the structure of the world and its affordances that enables prediction and planning. The existence of such models is fundamental to experience. As we move through the world, the raw instantaneous sensory information that we receive is highly impoverished and dynamic relative to the rich, organized, stable and detailed nature of experience. Learned and perhaps partially innate priors allow the maintenance of a veridical representation of the world around us, and rapid selection and integration of important information relevant to a task. In this proposal, we aim to probe the neural implementation of world models by recording from multiple brain areas in primates as they navigate naturalistic environments. Through modeling alongside analysis of coordinated recordings across multiple labs during a single sequence of complex tasks, we will develop a holistic understanding of how such models develop, depend on active engagement with the world, and influence perception. In our project, three macaque neurophysiology labs, with distinct cutting-edge expertise in different brain regions and technologies, will collaborate with computational neuroscientists with different expertise to pursue overlapping aims. We propose a novel collaboration strategy in which all three macaque neurophysiology labs will investigate a common navigation task, and record from different but overlapping subsets of areas: inferotemporal cortex (IT), motor cortex (MC) and prefrontal cortex (PFC). We focus initially on IT-HC (Tsao), MC-HC (Orsborne), and PFC-HC (Buffalo), on the hypothesis that HC, by virtue of its anatomy, its essential function in episodic memory formation, and its master role in orchestrating subjective experience, constitutes a central hub for representing the brain’s internal model of the world. To help to interpret this new data, we will develop and probe network models of predictive processing in collaboration with the analysis and theory team. Fairhall, Mihalas, Rao, and Shea-Brown, with complementary expertise in neural coding, predictive coding, and network dynamics, will interact to develop integrated model frameworks and work with each experimental lab individually on data analysis. Major questions we will address include: How does such a “world model” develop and how is it represented in the brain? What are the consequences of world models for the dynamics of sensory processing and behavior? How is learned structural knowledge integrated with self- motion during exploration of an environment? How is this information retrieved during memory- guided navigation to a goal?