Neural computation from retina to visual cortex

描述（由申请人提供）：这是一份关于计算神经科学合作研究的提案，由两个在视觉系统生物学（Meister）和统计与动态过程理论（Fisher）方面具有互补专业知识的研究小组组成。目的是更好地定量了解早期视觉系统的信息处理，从视网膜到初级视觉皮层。具体来说，该项目将为视网膜中的神经细胞如何对视觉输入作出反应提供改进的数学模型。然后，这些模型将用于探索人类视觉表现背后的后续大脑回路中神经计算的本质。这项工作的结果将导致更好地理解视觉，特别是区分视网膜和大脑的后续阶段在决定什么可以被感知方面所起的作用。这一基本认识将最终有助于识别视觉功能障碍。

DESCRIPTION (provided by applicant): This is a proposal for collaborative research in computational neuroscience, between two research groups having complementary expertise in the biology of the visual system (Meister) and the theory of statistical and dynamical processes (Fisher). The object is a better quantitative understanding of information processing in the early visual system, from retina to primary visual cortex. Specifically, the project will produce improved mathematical models for how nerve cells in the retina respond to visual input. Then those models will be used to explore the nature of neural computation in subsequent brain circuits that underlie human visual performance. Results from this work will lead to a better understanding of vision, specifically to distinguish what role the retina and subsequent stages of the brain play in determining what can be perceived. This basic understanding will eventually aid in the recognition of visual dysfunctions. The research will involve close interplay between experiment and theory. Visual responses will be recorded from ganglion cells of the isolated retina of salamander or rabbit, using a multi-electrode array. Dynamical systems models will be developed to emulate the relationship between the observed visual input and neural output. When two or more models appear plausible, suggestions for new experiments will emerge to apply additional tests. The goal of these efforts is a compact mathematical description of how the retina encodes a wide variety of stimuli. A second goal is to explore what computations might occur in subsequent circuits of the visual system. How can signals from different ganglion cells be combined to extract specific features, such as the precise location and trajectory of an object? The models of retinal function will be used to simulate spike trains that feed the lateral geniculate nucleus and primary visual cortex. The structure of those responses greatly constrains what and how the nervous system can compute with them, and different algorithms for combining retinal output signals predict very different performance. Comparison with the actual perceptual performance of animals and humans will distinguish between different hypotheses for computation in the early visual system.