论文部分内容阅读
光学高度互连可能是实际实施一种较新计算理论概念——神经网络的关键。加州理工学院 J.Hopfield 开创的神经网络概念的研究表明:神经的大量互连是大脑功能的关键要素。长期目标是建立能象人脑那样可以辩别图象的计算机。该领域的活动正在日益增大,87年3月16～18日召开的光计算机会议即是明证。加州理工学院 D.Psaltis 在一篇大会报告中说,“经由半导体技术进行综合联系也许是光学方法的最明显特征,光学神经计算机的发展可以看成开拓这一特征的尝试。在神经网络中,每一基本计算单元神经都直接与其它成千上万个单元联系,而电子计算机每个门一般只与2～3个门联系。借助于光学方法,有可能实现像神经网络那样高密度联系。”神经网络模型由神经和联接系统组成。神经作为阈元件,而信息存储在联接系统中。误差检测认为是对联接方式的修改,直至网络性能合理和没有差错为止。结构差异预计会导致光计算机与现代电子计算机很不相同。亚利桑那大学光学系统合作中心主任 H.M.Gibbs 说,
Optical height interconnection may be the key to actually implementing a newer concept of computational theory - neural networks. A study of the neural network concept pioneered by J. Hopfield at the California Institute of Technology has shown that the massive interconnection of nerves is a key element of brain function. The long-term goal is to create computers that can identify images like the human brain. The activities in this field are increasing day by day, and the light computer conference held from March 16 to March 18, 1987, is proof of this. In a conference report, D. Psaltis of the California Institute of Technology said, “Integrated connectivity via semiconductor technology may be the most obvious feature of optical methods and the development of optical neural computers can be viewed as an attempt to exploit this feature.” In neural networks , Each of the basic computational unit nerves is directly connected to tens of thousands of other units, whereas each door of an electronic computer generally only has 2 or 3 doors in contact with each other. With optical methods, it is possible to achieve high-density connections like neural networks . "The neural network model consists of a neural and a linked system. Nerves serve as threshold elements, and information is stored in the coupling system. Error detection is a modification of the connection method until the network performance is reasonable and no error. Structural differences are expected to result in very different optical computers from modern electronic computers. H.M. Gibbs, director of the Center for Optical Systems Collaboration at the University of Arizona,