本文提出了一类新的双向联想神经网络。与以往基于能量函数的神经网络不同,该网络通过在两个不同的Hilbert空间的交替投影进行联想,因而其动态特性和稳态解可以在信号空间上进行分析且具有快速自适应自学习等优点。该网络可做为异联想存贮器或分类器。本文给出了该网络在存贮和恢复二值模式时的实验结果,表明了其优良的性能。 This paper presents a new type of bi-directional associative neural network. Unlike previous energy-based neural networks, which are associated by alternating projections in two different Hilbert spaces, their dynamics and steady-state solutions can be analyzed in the signal space and have the advantages of fast adaptive self-learning . The network can be used as an associative memory or classifier. This paper gives the experimental results of the network in storage and recovery binary mode, indicating its excellent performance.