论文部分内容阅读
提出一种在大规模光电混合神经网络系统中实现可编程拓扑重构的有效方法。文中介绍了这种方法的原理、运用技术以及采用这种方法在已有ＮＰ１０２４光学／数字神经网络处理器中运行单层反馈型和多层前馈型等不同网络拓扑结构的实验结果
An efficient method to realize programmable topological reconstruction in a large-scale optoelectronic hybrid neural network system is proposed. This paper introduces the principle of this method, the application of technology and the experimental results of using this method to run different network topologies such as single-layer feedback and multi-layer feedforward in the existing NP1024 optical / digital neural network processor