arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2109.10951cs.NEcs.DBcs.DCq-bio.NC
收藏

一种人脑级神经网络的建模模式

Naming Schema for a Human Brain-Scale Neural Network

Morgan Schaefer, Lauren Michelin, Jeremy Kepner

摘要:

深度神经网络已经变得越来越大和稀疏,从而允许以降低的存储和计算成本来存储大规模神经网络。使用当前版本的高性能Apache Accumulo数据库和分布式维度数据模型(D4M)软件,可以存储与人脑连接一样多的神经网络。如此大规模的神经网络可能会引起人脑连接组社区科学家的特别兴趣。为了帮助研究和理解与现有神经网络(如大脑)并行的人工神经网络,可以开发命名模式以标记与大脑中的神经元组并行的人工网络中的神经元组。人工神经元组能够在小区域中进行特异性标记,以供未来研究。

英文摘要:

Deep neural networks have become increasingly large and sparse, allowing for the storage of large-scale neural networks with decreased costs of storage and computation. Storage of a neural network with as many connections as the human brain is possible with current versions of the high-performance Apache Accumulo database and the Distributed Dimensional Data Model (D4M) software. Neural networks of such large scale may be of particular interest to scientists within the human brain Connectome community. To aid in research and understanding of artificial neural networks that parallel existing neural networks like the brain, a naming schema can be developed to label groups of neurons in the artificial network that parallel those in the brain. Groups of artificial neurons are able to be specifically labeled in small regions for future study.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑