基于ART神经网络的在线学习分析策略算法研究  

Research of Online Learning Analysis Strategy Algorithm Based on ART Neural Network

作　　者：齐芳[1] 陈静[1] 赵培军[1] 

机构地区：[1]天津体育学院,天津300381

出　　处：《激光杂志》2016年第7期118-123,共6页Laser Journal

基　　金：天津科技计划项目(14ZCDGSF0004);天津市教委科研计划项目(20132520);天津市普通高校教改研究计划项目(C03-0809)

摘　　要：本文在分析了大数据时代在线学习策略的必要性及其面临的挑战的前提下,设计了基于ART神经网络的在线学习分析策略算法,ART神经网络采用以认知和行为模式为基础的无指导的竞争学习算法实现矢量聚类,良好地维持了网络鲁棒性和可调节性的平衡关系,既能够非常灵活的适应新型输入模式产生的变化,同时又可以避免对网络已记忆模式进行篡改。基于ART神经网络模型的在线学习分析策略主要包括识别阶段、比较阶段、学习阶段以及搜索阶段。通过实验设定输入模式,完成在线学习过程进行测试验证。实验结果表明,本文提出的基于ART神经网络的在线学习分析策略算法能够实现对不同输入模式较高正确率的分类,同时可通过警戒参数对网络的分类粒度进行调节,使在线学习策略具有自适应性。Through analyzing the necessity of online learning strategy under the premise of the challenge in the era of big data, the online learning analysis strategy based on ART neural network algorithm was designed. ART neural network ase competitive learning algorithm on the basis of the cognitive and behavioral mode without guiding to realize vector clustering which can better solve the relationship between the network stability and plasticity, and it cannot only has great flexibility for adapting to the new input model, but also can avoid modifying the previous learning mode of the network. The online learning strategy based on ART neural network model mainly includes the recognition stage, comparison stage, learning phase and search phase. The input mode is set through the experiment, and then the validation of online learning process was completed. The experimental results show that the online learning analysis strategy based on ART neural network algorithm can realize classification with high accuracy according input mode, at the same time network classification granularity can be adjusted through vigilance parameters, which makes the learning strategy having adaptability.

关 键 词：在线学习 ART神经网络 学习规则 竞争学习 

分 类 号：TN249[电子电信—物理电子学]