鲁棒性增强的ART1神经网络

IMPROVEMENT OF ART1 NEURAL ARCHITECTURE IN ROBUSTNESS

本文详细讨论了ＡＲＴ１神经网络实现算法过程中匹配因子的选择，提出了适用于消除信号噪声以及恢复信号中部份缺损的信息的匹配因子．从而提高了ＡＲＴ１在进行模式分类的工作性能，改善了ＡＲＴ１抗噪性能，一定程度实现了从部份信息丢失的信号中联想恢复出完整信号，增强了ＡＲＴ１的鲁棒性．

In this paper we consider the original model of ART1 introduced by. Carpenter and Grossberg especially in network algorithm. A new ART1-based neurel architecture is presented in order to improve the robustness. The underlying theory and the improvement on the ART1 model are studied for classifying optical character patterns in the presence of noise or lost parts of pixels without prior knowledge. The experimental results show that our new ART1 netal networks have performed well and coincided with our analysis.