Ultra parallel computation using wave interference as means of weighted sum : Holographic neural computing

使用波干涉作为加权和的超并行计算：全息神经计算

• 批准号: 10680339
• 负责人: KUMAZAWA Itsuo
• 金额: $ 1.98万
• 依托单位: Tokyo Institute of Technology
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1998
• 资助国家: 日本
• 起止时间: 1998 至 1999
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-10680339/
• 关键词: Holographic Memory; Neural Network; Laser; Coherent wave; Parallel computation; Wave interference; Diffraction; Ultrasonic wave; ニュートラルネットワーク; 回折格子; 光コンピュータ; 超並列計算機; ホログラフィー; 波動; 位相; 画像処理

he project developed a novel approach for ultra parallel computation using interference of coherent waves as direct means of weighted snm computation for artificial neural networks. Coherent waves such as acoustic waves and lasers can interfere with each other and the amplitude of synthesized wave is determined by their phase differences. This phenomena can be utilized as means of weighted sum computation. Weighted sum is a basic operation of a neuron in the artificial neural networks which have been extensively studied for artificial intelligence purposes. The developed system uses only wave emitting and receiving devices as its components. The relative positions of these components determines the phase differences of the waves and, consequently, the ways of weighted sum. The kind of computation is coded as the positions of these components arranged on a two dimensional surface. In this fashion, the system does not require neither communication wires nor computation devices such as adder and multiplier. During the two years of this project, we proposed several ways of implementing this idea and examined the computational capabilities of the proposed systems under noisy and distorted wave conditions.

该项目开发了一种新的超并行计算方法，利用相干波的干涉作为人工神经网络加权SNM计算的直接手段。相干波如声波和激光可以相互干扰，合成波的振幅由它们的相位差决定。这种现象可以作为加权和计算的手段。加权和是人工神经网络中神经元的一种基本运算，在人工智能领域得到了广泛的研究。所开发的系统仅使用波发射和接收器件作为其组成部分。这些分量的相对位置决定了波的相位差，从而决定了加权和的方法。这种计算被编码为这些组件在二维表面上的位置。以这种方式，系统既不需要通信线路，也不需要加法器和乘法器等计算设备。在这个项目的两年中，我们提出了几种实现这一想法的方法，并检查了所提出的系统在噪声和畸变波条件下的计算能力。

项目成果

I. Kumazawa and Y. Kure: "Computation of Weighted Sum by Physical Wave Properties - Coding Problems by Unit Positions"In George D. Smith, Nigel C. Steele and Rudolf F. Albecht (eds.) Artificial Neural Nets and Genetic Algorithms. Springer Computer Science

I. Kumazawa 和 Y. Kure：“通过物理波特性计算加权和 - 按单位位置编码问题”，乔治 D. 史密斯、奈杰尔 C. 斯蒂尔和鲁道夫 F. 阿尔贝希特（编辑）人工神经网络和遗传算法。

Itsuo Kumazawa (D. Smith(eds.)): "Artificial Neural Nets and Genetic Algorithms"Springer Computer Science. 634 (1998)

Ituo Kumazawa（D. Smith（编辑））：“人工神经网络和遗传算法”施普林格计算机科学。

I. Kumazawa: "A Cellular neural network framework for shape representation and matching"Proceedings of Third International Conference on Knowledge-based Intelligent Information Engineering Systems. IEEE. 178-181 (1999)

I. Kumazawa：“用于形状表示和匹配的细胞神经网络框架”第三届基于知识的智能信息工程系统国际会议论文集。

石井真樹: "重み表現に線形の従属制約を導入した階層型ニューラルネットワークの写像能力"電子情報通信学会ニューロコンピューティング研究会. NC99-36. 57-62 (1999)

Maki Ishii：“引入权重表示的线性依赖约束的分层神经网络的映射能力”IEICE NC99-36 (1999)。

Itsuo Kumazawa: "Learning and tracking target shapes by compact neural networks"Proceedings of ICONIP/AMZIIS/ANNES'99 International Workshop. 41-44 (1999)

Ituo Kumazawa：“通过紧凑神经网络学习和跟踪目标形状”ICONIP/AMZIIS/ANNES99 国际研讨会论文集。