Neural Computation in the Retina

描述(由申请人提供)：本项目的主要目标是了解神经电路可以执行的计算。脊椎动物视网膜被选为一个模型系统，因为我们有能力实现它的输出的完整记录，同时用现实的光模式刺激，以及它复杂和良好的解剖结构。视网膜对视觉刺激的表示在光感受器和神经节细胞之间发生了巨大变化，前者使用类似相机的表示，其中每个细胞是一个像素，后者表示具有复杂且高度重叠的视觉特征的基本集合的图像。我们试图用视网膜神经节细胞的完整群体来描述视觉场景的特征。 具体目标是：1)通过进行功能分类来表征神经节细胞群体中所代表的视觉特征的多样性和刻板印象；2)测量平行视觉通道内和平行视觉通道之间的相关性；3)制定解码策略，以利用大量神经节细胞群体进行视觉辨别。我们的功能分类将通过同时记录来自单个视网膜的200多个神经节细胞来最大限度地减少不同视网膜制剂之间的细微差异的影响，并将使用信息论来使用包括自然电影在内的一系列视觉刺激来评估功能差异。我们对神经节细胞组之间的相关模式的描述将使用新的信息论技术，包括最大熵方法。最后，我们的解码器将寻求利用我们在大量人群中的相关模型来提高视觉辨别的准确性。

DESCRIPTION (provided by applicant): The broad goals of this project are to understand the computations that can be performed by a neural circuit. The vertebrate retina is chosen as a model system, because of our ability to achieve complete recordings of its output while stimulating with realistic patterns of light, as well as its intricate and well-characterized anatomy. The retina's representation of visual stimuli changes dramatically between the photoreceptors, which use a camera-like representation where each cell is a pixel, and the ganglion cells, which represent the image with a complex and highly overlapping basis set of visual features. We seek to characterize the manner in which visual scenes are represented by the complete population of retinal ganglion cells. The Specific Aims are: 1) characterize the diversity and stereotypy of the visual features represented within the population of ganglion cells by performing functional classification; 2) measure the correlation both within and among the parallel visual channels; 3) formulate decoding strategies to perform visual discrimination using large populations of ganglion cells. Our functional classification will minimize the effects of subtle variations from one retinal preparation to the next by recording simultaneously from 200+ ganglion cells from a single retina and will use information theory to assess functional differences using a battery of visual stimuli including natural movies. Our characterization of the patterns of correlation among groups of ganglion cells will use new information-theoretic techniques, including the maximum entropy approach. Finally, our decoders will seek to make use of our models of correlation in large populations to improve the accuracy of visual discriminations.