首页>
外国专利>
METHOD AND DEVICE FOR SIMULATING NEURAL NETWORK COMPLETELY IN PARALLEL BY USING DIGITAL INTEGRATED CIRCUIT
METHOD AND DEVICE FOR SIMULATING NEURAL NETWORK COMPLETELY IN PARALLEL BY USING DIGITAL INTEGRATED CIRCUIT
展开▼
机译:用数字集成电路并行仿真神经网络的方法和装置
展开▼
页面导航
摘要
著录项
相似文献
摘要
PURPOSE: To speed up the learning of a multi-layered neural network by performing calculation for correcting synapsis weight in parallel and also carrying out calculation for correcting a virtual target in parallel. ;CONSTITUTION: A neuron inputs an input pattern Ok from a precedent layer, multiplies a presynapsis virtual target vector Tk by a synapsis weight vector Wk, and performs the sigmoid transformation of their sum to calculate an input Ok+1 to the following layer. Here, (k) is the number of layers. Then the weight Wk is calculated by using an error between the input Ok+1 to the following layer and the virtual target Tk+1 of the following layer and the virtual target Tk to be updated. Then the virtual target Tk is calculated by using presynapsis and postsynapsis virtual targets Tk and Tk+1, presynapsis and postsynapsis input patterns Ok and Ok+1, and the updated weight Wk and updated. The calculating processes in the respective steps are all performed in parallel.;COPYRIGHT: (C)1994,JPO&Japio
展开▼
机译:目的:通过并行执行校正突触权重的计算并并行执行校正虚拟目标的计算,来加快多层神经网络的学习。 ;构成:神经元从先例层输入输入模式O k Sup>,将突触前虚拟目标向量T k Sup>乘以突触权重向量W k Sup>,并对它们的和执行S型变换以计算到下一层的输入O k + 1 Sup>。在此,(k)是层数。然后,使用到下一层的输入O k + 1 Sup>与虚拟目标T k + 1 Sup之间的误差来计算权重W k Sup>下一层的>和要更新的虚拟目标T k Sup>。然后使用突触前和突触后虚拟目标T k Sup>和T k + 1 Sup>,突触前和突触后输入模式O计算虚拟目标T k Sup> k Sup>和O k + 1 Sup>,更新后的权重W k Sup>并更新。各个步骤中的计算过程都是并行进行的。版权所有:(C)1994,日本特许厅&日本apio
展开▼