光学学报, 1996, 16 (8): 1128, 网络出版: 2006-12-04
基于正交化算法的三值互连神经网络模型
A Neural Networks Model with Trinary Interconnection Weights Based on Orthogonalized Algorithm
摘要
提出了以信息损失最少为原则的三值(±1)互连权重编码方法,这种编码方法比以前的三值权重编码方法显著地提高了神经网络的性能。由于互连权重只有三值,恰恰弥补了光互连精度不高的缺点,易于光学实现。
Abstract
An encoding method for constructing trinary interconnection weights based on the rule of least-lost of information is proposed in this paper. The encoding method improves the limitation on the accuracy of optical interconnection and is potential for optical implementation. The performance of ONN is evidently improved by this encoding method compared to the other methods proposed before. A detailed description of this method is presented.
常胜江, 杨建文, 高胜泉, 申金媛, 张延炘. 基于正交化算法的三值互连神经网络模型[J]. 光学学报, 1996, 16(8): 1128. 常胜江, 杨建文, 高胜泉, 申金媛, 张延炘. A Neural Networks Model with Trinary Interconnection Weights Based on Orthogonalized Algorithm[J]. Acta Optica Sinica, 1996, 16(8): 1128.