Mechanisms of efficient coding of dynamic visual motion signals for pursuit

描述（由申请人提供）：人眼向大脑发送信息的速度估计约为每秒10兆比特，大致相当于以太网连接的速度。在行为相关的时间尺度上处理如此大带宽的视觉信息流要求神经元有效地从视觉信号中提取和表示信息，即以最小的时间和能量代价表示最多的信息。从本质上讲，大脑需要压缩视觉流，就像软件压缩电影的数字表现一样。人们对大脑如何完成这项关键任务知之甚少。我们建议研究层外视觉皮层在单个神经元、群体和追求眼球运动行为中用于编码运动信息的神经机制。脑中脑区的神经元选择性地响应视觉运动，并为眼球的平滑运动提供视觉输入。通过记录神经和行为反应，我们不仅可以确定皮层神经元如何压缩输入的视觉信号以有效地表示它们，还可以确定这些编码策略对行为表现是否重要。我们将以该范式为基础，研究MT神经元如何联合编码运动信息，并以最近在视网膜上的工作为指导，证明神经群对刺激的压缩增强。本研究的总体目标是确定动态视觉运动刺激是如何在皮层神经元群中表现出来的，以及这些感官信息随后被读出以产生追求的效率如何。长期目标是确定在自然条件下，大脑是如何表现动态感官信息和解码行为线索的。这个项目可能会对我们理解大脑在自然条件下如何处理刺激，以及我们如何概念化感官处理产生深远的影响。这项研究将有助于视网膜修复软件的开发，通过阐明大脑如何对运动场景进行编码来弥补中央视觉处理的缺陷。本研究的目的是研究(1)动态运动刺激在脑皮层的有效感觉编码。一种自适应的感觉编码通过调整灵敏度和对当前刺激条件的整合时间来最大化信息传递。我们将测试MT神经元自适应编码运动的假设，以及它们的编码效率影响眼睛飞行时的追踪性能。在自然运动的场景中，运动的波动在许多时间尺度上是相关的。我们将通过计算当前响应和未来刺激之间的互信息来衡量MT发射和自然运动跟踪的压缩效率。我们的第二个目标(2)是量化追踪运动输入的动态MT总体编码。我们将使用追踪的精度作为约束皮质群体编码模型的基准，剖析峰值和沉默模式的贡献-跨越时间和跨MT神经元群体-对目标运动的编码并测量编码池的大小。

DESCRIPTION (provided by applicant): The human eye sends information to the brain at an estimated rate of about 10 megabits per second, roughly the speed of an ethernet connection. Processing such a large bandwidth stream of visual information on behaviorally relevant time scales requires that neurons extract and represent information from visual signals efficiently, i.e represent the most information for the least cost in time and energy. In essence, the brain needs to compress the visual stream much the same way software compresses the digital representation of a movie. Little is known about how the brain accomplishes this critical task. We propose to investigate the neural mechanisms that extrastriate visual cortex uses to encode motion information in single neurons, populations, and in pursuit eye movement behavior. Neurons in area MT respond selectively to visual motion and provide the visual inputs for smooth pursuit eye movements. By recording neural and behavioral responses together, we can determine not only how cortical neurons compress incoming visual signals to represent them efficiently but also whether those coding strategies are important for behavioral performance. We will build on that paradigm to study how MT neurons jointly encode motion information, guided by recent work in the retina demonstrating enhanced stimulus compression by neural populations. The general aim of the proposed research is to determine how dynamic visual motion stimuli are represented in a cortical neuronal population and how efficiently that sensory information is subsequently read out to generate pursuit. The long-term goal is to determine how the brain represents dynamic sensory information and decodes the cues for behavior under natural conditions. This project could have a profound impact on our understanding of how the brain processes stimuli under natural conditions and for how we conceptualize sensory processing. The study will aid the development of software for retinal prosthetics that will remediate deficits in central visual processing by elucidating how the brain encodes moving scenes. Our Aims are to study (1) the Efficient sensory coding of dynamic motion stimuli in cortical area MT and pursuit. An adaptive sensory code maximizes information transfer by adjusting sensitivity and integration time to the current stimulus conditions. We will test the hypothesis that MT neurons adaptively encode motion, and that their coding efficiency impacts pursuit performance while the eyes are in flight. In natural moving scenes, fluctuations in motion are correlated across many time scales. We will measure the compression efficiency in MT firing and pursuit tracking of naturalistic motion by computing the mutual information between present response and future stimulus. Our second Aim (2) is to Quantify dynamic MT population encoding of motion inputs for pursuit. We will use the precision of pursuit as a benchmark to constrain models of cortical population coding, dissecting the contribution of patterns of spikes and silences -- across time and across populations of MT neurons -- to the encoding of target motion and to measure the size of the coding pool.