STATISTICAL APPROACHES TO UNDERSTANDING NEURAL FEATURE SELECTIVITY

这个子项目是许多研究子项目中的一个 由NIH/NCRR资助的中心赠款提供的资源。子项目和 研究者（PI）可能从另一个NIH来源获得了主要资金， 因此可以在其他CRISP条目中表示。所列机构为 研究中心，而研究中心不一定是研究者所在的机构。 这个项目的目标是研究在几个神经元群体中的神经元进行的信号处理，这些神经元不屈服于线性系统分析。通常，皮层神经元需要多个滤波器来描述其编码，并对滤波输入进行各种非线性运算。Teragrid资源将用于为神经元找到两个最相关的滤波器：（1）初级听觉皮层中的神经元，目的是表征发生在六个皮层层中的信号处理的多样性;（2）鸣禽听觉前脑中的神经元，以确定其频率和时间曲线;（3）在灵长类动物外侧膝状体中，这是一个既可以分析输入又可以分析输出的系统，因此我们可以研究单个神经元在滤波中的转换;（4）啮齿类动物的视皮层与其他物种相比，总体上描述较少。通过最大化神经响应序列和由它们过滤的输入刺激之间的信息香农互信息来找到相关的神经过滤器（Sharpee等人，神经计算2004）。我们采用了C++代码与开放MP并行化，并遵循模拟退火计划相结合的搜索沿着梯度的信息。该任务非常适合并行计算，因为每个神经元的计算彼此独立。然后使用本地资源对所获得的相关神经拟合器进行可视化和分析，以提供对上述神经元群体的典型特征选择性和组织的见解

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. The goal of this project are to study signal processing carried out by neuron in several populations of neurons which do not yield to linear systems analysis. Quite often cortical neurons require multiple filters to describe their coding, with diverse nonlinear operations on the filtered inputs. The Teragrid resource will be used to find the two most relevant filters for neurons (1) in the primary auditory cortex with the goal of characterizing the diversity of signal processing that takes place across the six cortical layers; (2) in the auditory forebrain of songbirds in order to determine their profile in both frequency and time; (3) in the lateral geniculate nucleus of primates where, which is a system where both inputs and outputs to the circuit can be analyzed, so taht we can study the transformation in filtering carried out by single neurons; (4) the rodent visual cortex that is overall relativel less described compraed to other species. The relevant neural filters are found by maximizing information Shannon muutal information between the sequence of neural responses and input stimuli filtered by them (Sharpee et al Neural Computation 2004). We employ a C++ code with open MP parallelization, and follow a simulated annealing scheme combined with a search along the gradient of information. The task is well suited for parallel computation, because calculations for each neuron are independent of each other. The obtained relevant neural fitlers are then visualized and analyzed using local resources to provide insights into the typical feature selectivity and organization of the populations of neurons listed above