Parallel Processing in a Neural Network

神经网络中的并行处理拟议研究的目标是进一步深入了解神经系统中的并行处理机制。研究将集中在前庭-眼睛反射(VOR)，它通过产生眼睛旋转来抵消头部旋转来稳定视力。VOR是由前庭核神经元(VNN)介导的。最近的神经网络模型表明，由于VNN是并行操作的，所以对眼动命令进行分布式编码的结果是VNN。基本的假设是并行性对VOR函数至关重要。该项目的主要目的是通过研究不同数量的VNN对VOR的影响来定量检验这一假设。实验对象将是金鱼，因为它们有适当发育良好的VOR。为了生成预测，将从高度平行和真实的金鱼VOR模型中移除不同数量的VNN。初步预测，VNN病变应通过增加VOR谐波失真来降低VOR功能。这将通过比较VNN损伤前后的VOR眼球运动来进行实验验证。数据将使用眼速功率谱进行量化。病变大小将在组织学上进行量化。结果将被合并到更大、更现实的VOR神经网络模型中，该模型将在大规模并行计算机上实施。//

PARALLEL PROCESSING IN A NEURAL NETWORK The goal of the proposed research is to gain further insight into mechanisms of parallel processing in the nervous system. Research will focus on the vestibulo-ocular reflex (VOR), which stabilizes vision by producing eye rotations that counterbalance head rotations. The VOR is mediated by vestibular nuclei neurons (VNNs). Recent neural network models suggest that the distributed encoding of eye-movement commands VNNs results because VNNs operate in parallel. The underlying assumption is that parallelism is critical to VOR function. The primary aim of the project is to quantitatively test this assumption by studying the effects on VOR of lesioning varying numbers of VNNs. Experimental subjects will be goldfish because they have a suitably well-developed VOR. To generate predictions, varying numbers of VNNs will be removed from a highly parallel and realistic model of the goldfish VOR. The initial prediction is that VNN lesions should degrade VOR function by increasing VOR harmonic distortion. This will be tested experimentally by comparing VOR eye-movements before and after VNN lesions. Data will be quantified using eye-velocity power spectra. Lesion size will be quantified histologically. Results will be incorporated into larger and more realistic VOR neural network models, which will be implemented on massively parallel computers. //