Probing neural feature selectivity with natural stimuli

用自然刺激探测神经特征选择性

基本信息

批准号：
7418710
负责人：
Tatyana O. SHARPEE
金额：
$ 9.78万
依托单位：
SALK INSTITUTE FOR BIOLOGICAL STUDIES
依托单位国家：
美国
项目类别：
财政年份：
2004
资助国家：
美国
起止时间：
2004-04-07 至 2010-03-31
项目状态：
已结题

项目摘要

DESCRIPTION (provided by applicant): The candidate seeks to become independent as a neuroscientist and to work at the interface between neuroscience and physics. The Sloan-Swartz Center at UCSF provides a unique opportunity to become immersed in neuroscience at its cutting edge. While the majority of her time will be spent in research, she will attend a range of systems neuroscience and computer vision courses. The immediate goals of her research are to develop an information-theoretic method that allows for a rigorous statistical analysis of neural responses to natural stimuli, which are non- Gaussian and have strong spatiotemporal correlations; and to use this method to analyze the responses of visual cortical neurons to natural time-varying images. There are numerous indications that the responses of neurons to natural stimuli cannot be completely predicted from their responses to stimuli with simple statistical properties, such as white noise ensembles. However, existing methods for analyzing single neuron responses may be rigorously applied only to Gaussian ensembles. The information-theoretic method involves finding the stimulus dimensions that carry the most information about the neuron's response. The stimulus ensemble is not assumed to be Gaussian. The only assumption made is that the neuron is selective for a small number of stimulus dimensions out of the high-dimensional stimulus space: responses with respect to the relevant dimensions might be arbitrarily nonlinear. After testing the method on model neurons and verifying that it gives reasonable results for cortical neurons probed by natural scenes, the method will be used to test the categorical distinction between simple and complex cells in the primary visual cortex. The method will then be modified so that a large set of directions related to each other via certain symmetry operations, such as translation or scaling, can be simultaneously found, with the goal of systematically probing neurons in extrastriate areas with natural scenes.

描述（由申请人提供）：候选人寻求独立作为神经科学家，并在神经科学和物理学之间的界面上工作。 UCSF的Sloan-Swartz中心提供了一个独特的机会，可以沉浸在其最前沿的神经科学中。尽管她的大部分时间都将用于研究，但她将参加一系列系统神经科学和计算机视觉课程。她的研究的直接目标是开发一种信息理论方法，该方法允许对自然刺激的神经反应进行严格的统计分析，该反应是非高斯，并且具有很强的时空相关性；并使用这种方法来分析视觉皮质神经元对自然时变图像的响应。有许多迹象表明，神经元对自然刺激的反应不能完全从它们对刺激的反应中完全预测，而这些刺激具有简单的统计特性，例如白噪声集合。但是，现有的分析单个神经元反应的方法只能严格应用于高斯集合。信息理论方法涉及找到有关神经元反应的最多信息的刺激维度。刺激合奏不被认为是高斯。唯一的假设是神经元对高维刺激空间中的少量刺激维度具有选择性：相对于相关维度的响应可能是任意非线性的。在测试模型神经元上的方法并验证它为自然场景探测的皮质神经元提供了合理的结果后，该方法将用于测试主要视觉皮层中简单和复杂细胞之间的分类区别。然后将修改该方法，以便可以通过某些对称操作（例如翻译或缩放）相互相互关联的大量方向，以便在具有自然场景的外部区域系统地系统地探测神经元。