CRCNS: Coding for optimal performances in natural environments

描述(由申请人提供)：在神经编码中捕获自然界的统计结构对于最佳适应环境是必不可少的。这项提案通过询问大脑如何在谷仓猫头鹰的声音定位系统中接近统计最优来研究这个问题。贝叶斯理论框架将被用来描述如何将感觉和先验信息最佳地结合起来，以指导定向行为。具体地说，我们试图证明，感觉可靠性和先验信息表示在代表听觉空间的神经群体的响应属性和地形图中。第一个目标是研究感觉线索可靠性是如何在大脑中表现出来的。感觉信息的最佳使用要求感觉线索的统计可靠性可以从神经反应中获得。以前的理论认为，线索可靠性编码在神经反应的增益中，或者神经反应的选择性中，但可靠性如何表示尚不清楚。在OWL中，空间线索统计可靠性的变化导致声音定位行为的变化，符合贝叶斯模型。我们的模型预测可靠性编码在空间特定神经元的调谐曲线宽度中 猫头鹰的中脑。我们将独立地操纵调谐曲线宽度和射击率来检验这一假设，并用行为来检验模型。第二个目标将研究声音定位的空间线索整合是否遵循统计最优规则。自然环境中的感知往往依赖于多个线索的整合，无论是在通道内还是跨通道。这里，积分是线性的还是非线性的是至关重要的，因为将贝叶斯模型从一维扩展到二维表明，条件独立的感觉线索的最佳组合应该是非线性的。在猫头鹰的大脑中，用于确定仰角和方位的空间线索是独立处理的，并在中脑以非线性方式组合形成空间感受野。然而，声音定位线索是否具有条件性独立还不得而知。这一目标将说明为什么非线性运算对于最佳线索组合是必不可少的，以及它们是如何产生的。我们将进行体内细胞内记录和行为测试来解决这些问题。这将为条件独立线索的最佳组合是非线性的预测提供实验检验。第三个目标是将该模型扩展到动态听觉场景的编码中；将时间维度融入到声音定位的贝叶斯模型中。我们将使用种群向量模型来确定神经系统如何通过贝叶斯推理在听觉空间实现预测能力。我们将测量中脑神经元在空间和时间上的感受野，以检验猫头鹰对朝向凝视中心的来源有偏见的假设。我们将使用行为测试来测量运动声源的检测阈值。最后，我们将研究非均匀网络中的动态增益控制是否可以解释具有偏向凝视中心的声源的贝叶斯预测编码。更广泛的影响：关于如何通过神经编码捕获自然场景的统计数据的悬而未决的问题包括如何表示感觉信息的可靠性并将其与先前的概率知识相结合，以及如何整合感觉线索以最佳地指导行为。这个项目在猫头鹰听觉中脑空间的异质表示中解决了这些问题。是否可以使用种群向量来解码非均匀表示来执行贝叶斯推理，以及这种机制在多个维度上的工作超过了谷仓猫头鹰的声音定位，成为神经编码的普遍兴趣。参与这个项目的PI，其中一位是初级研究员，收集了建模、生理学和行为方法方面的互补专业知识，从而实现了真正的跨学科方法。因此，该项目将巩固强大的合作，同时提供有关神经科学悬而未决问题的开创性信息。参与其中的三个机构致力于培训任职人数不足的群体。阿尔伯特·爱因斯坦医学院位于布朗克斯区，使其成为美国最多元化和最贫穷的县之一的发展极，并为直接获得转化研究提供了可能性。西雅图大学数学系的加入将确保该项目将加强从本科生到博士后的培训。西雅图大学是西方十大本科课程之一，俄勒冈大学也是其中之一。

DESCRIPTION (provided by applicant): Capturing nature's statistical structure in the neural coding is essential for optimal adaptation to the environment. This proposal investigates this issue by asking how the brain can approach statistical optimality in the sound localization system of barn owls. A Bayesian theoretical framework will be used to describe how sensory and a priori information can be combined optimally to guide orienting behavior. Specifically, we seek to demonstrate that sensory reliability and a priori information are represented in the response properties and topography of the neural population that represents auditory space. The first aim studies how sensory cue reliability is represented in the brain. Optimal use of sensory information requires that the statistical reliability of sensory cues is accessible from neural responses. Previous theories have suggested that cue reliability is encoded in the gain of neural responses or alternatively the selectivity of neural responses but how reliability is represented is not known. In the owl, changes in the statistical reliability of spatial cues resultin changes in sound localization behavior consistent with a Bayesian model. Our model predicts that the reliability is encoded in the tuning curve widths of space-specific neurons located in the owl's midbrain. We will manipulate tuning-curve widths and firing rates independently to test this hypothesis and test the model with behavior. The second aim will study whether the integration of spatial cues for sound localization follows the rules of statistical optimality. Perception in natural environments often depends on the integration of multiple cues, both within modalities and across modalities. Here, whether the integration is linear or nonlinear is crucial, as extending a Bayesian model from one to two dimensions indicates that optimal combination of conditionally independent sensory cues should be nonlinear. In the owl's brain, the spatial cues used to determine elevation and azimuth are processed independently and combined nonlinearly in the midbrain to form spatial receptive fields. However, whether or not sound localization cues are conditionally independent is unknown. This aim will demonstrate why nonlinear operations are essential for optimal cue combination and how they arise. We will perform in vivo intracellular recording and behavioral tests to address these questions. This will provide an experimental test of the prediction that optimal combination of conditionally independent cues is nonlinear. The third aim will extend the model to coding dynamic auditory scenes; the time dimension will be incorporated into the Bayesian model of sound localization. We will use a population vector model to determine how a neural system can achieve predictive power in auditory space through Bayesian inference. We will measure receptive fields of midbrain neurons in space and time to test the hypothesis that the owl has a bias for sources moving toward the center of gaze. We will use behavioral tests to measure detection thresholds for moving sound sources. Finally, we will study whether a dynamic gain control in a non-uniform network can account for Bayesian predictive coding of sound motion with a bias for sources moving toward the center of gaze. Broader Impacts: Outstanding open questions of how statistics of natural scenes are captured by neural coding include how reliability of sensory information is represented and combined with prior probabilistic knowledge, and how sensory cues are integrated to optimally guide behavior. This project addresses these questions in the heterogeneous representation of space of the owl's auditory midbrain. Whether non-uniform representations can be decoded using a population vector to perform Bayesian inference and that this mechanism works in multiple dimensions transcends sound localization in barn owls, becoming of general interest to neural coding. The PIs involved in this project, one of them a junior researcher, gather complementary expertise in modeling, physiology and behavioral approaches allowing for a truly interdisciplinary approach. This project will thus consolidate a powerful collaboration while providing groundbreaking information on outstanding questions in Neuroscience. The three institutions involved are committed to the training of underrepresented groups. The location of the Albert Einstein College of Medicine in the Bronx, makes it a pole of development in one of the most diverse and poor counties in the country and provides the potential for direct access to translational research. The inclusion of the Department of Mathematics at Seattle University, ranked among the top ten universities in the West for undergraduate programs, and the University of Oregon will ensure that this project will enhance training from the undergraduate to postdoctoral levels.