Project C: Neural basis of causal inference in continuous navigation

项目摘要 当感觉输入不明确时，大脑建立一个内部模型来推断世界上的哪些事件 导致了这种感觉活动的模式。这个过程称为因果推理，提供了一个统一的框架 了解代表对世界结构的信念的神经信号如何与传入的 感官信号驱动感知-动作循环。这项建议关注的是 物体运动、物体深度和动物在整个世界中的自我运动，作为一种特殊的运动模式 视网膜上的神经活动可以由世界上物体的运动和自我运动的许多组合产生 动议。总体假设是，顶叶和前额叶神经元可以推断物体是否在 这些信号流经反馈投影以更新中与任务相关的表示 纹状外视皮质。这个项目的目标是研究动态任务中的因果推理，其中 动物对世界的内在模型不断变化。在猴子的虚拟现实导航任务中 这些实验将探索在自然环境下的大脑计算和多区域相互作用 持续行动和主动感知，以及对潜在的任务相关变量的动态在线推理 与内部模型相关。这个项目将开发一个动态导航任务的因果推理版本 已经在Angelaki实验室中使用，然后使用种群记录和因果神经操作 测试和改进A项目理论团队开发的动态模型。连续时间潜伏 该模型的变量将与猴子和老鼠的行为数据进行拟合，以揭示每种动物对 世界状态和相互作用的任务相关变量，并生成关于 神经动力学。利用多电极记录以及化学和光遗传操作，该项目 将在四个相互关联的猴子大脑区域中测试这些假设，这些区域涉及视觉感知， 导航、记忆和决策：顶叶7a区、前额叶8av区和纹外视皮层 MSTd区(内侧上颞区)和MT区(中颞区)。最后，神经活动将是 在整个小鼠大脑中绘制，重点是皮质下结构，使用平行记录和 神经像素探测器用于无假设地识别受该动态任务调制的其他区域， 这也将有助于将这些发现推广到不同物种。基于这些发现，更多的猕猴 根据需要，大脑区域将被用于记录和操作实验。总而言之，这些 实验将严格测试跨物种的动态因果推理的计算框架 大脑区域。与项目B中以试验为基础的任务的补充结果相比，成功 这些实验的完成有望揭示因果推理功能的一般原理 在自然主义和动态流动行为期间的过程和自上而下的反馈连接。

Project Summary When sensory inputs are ambiguous, the brain builds an internal model to infer which events in the world caused this pattern of sensory activity. This process, called causal inference, provides a unifying framework for understanding how neural signals that represent beliefs about the structure of the world interact with incoming sensory signals to drive perception-action loops. This proposal focuses on perceptual interactions among object motion, object depth, and an animal's self-motion through the world, as a particular moving pattern of neural activity on the retina can be generated by many combinations of object motion in the world and self- motion. The overall hypothesis is that parietal and prefrontal neurons infer whether an object moves in the world, and that these signals flow through feedback projections to update task-relevant representations in extrastriate visual cortex. The goal of this project is to study causal inference in dynamic tasks, in which an animal's internal model of the world changes continuously. In a virtual reality navigation task in monkeys and mice, these experiments will explore brain computation and multi-area interactions in the naturalistic setting of continuous action and active sensing, as well as dynamic on-line inference about latent, task-relevant variables related to the internal model. This project will develop a causal inference version of a dynamic navigation task already in use in the Angelaki laboratory and then use population recordings and causal neural manipulations to test and refine the dynamic model developed by the theory team in Project A. The continuous-time latent variables of this model will be fitted to monkey and mouse behavioral data to reveal each animal's beliefs about the state of the world and interacting task-relevant variables, and to generate novel hypotheses about the neural dynamics. Using multi-electrode recordings and chemical and optogenetic manipulations, this project will test these hypotheses in four mutually interconnected monkey brain areas involved in visual perception, navigation, memory, and decision-making: parietal area 7a, prefrontal area 8aV, and extrastriate visual cortical areas MSTd (dorsal medial superior temporal) and MT (middle temporal). Finally, neural activity will be mapped throughout the mouse brain, with an emphasis on subcortical structures, using parallel recordings with Neuropixels probes for hypothesis-free identification of other areas that are modulated by this dynamic task, which will also serve to generalize the findings across species. Based on these findings, additional macaque brain regions will be targeted for recording and manipulation experiments as needed. Collectively, these experiments will rigorously test the computational framework of dynamic causal inference across species and brain areas. When compared with the complementary findings from trial-based tasks in Project B, successful completion of these experiments is expected to uncover general principles of the function of causal inference processes and top-down feedback connections during naturalistic and dynamically fluid behavior.