摘要
分析了数据挖掘领域面临的性能问题(主要包括算法的有效性、可伸缩性和并行性);根据数据并行的思想,提出了在时序预测中并行训练神经网络的模型,以提高训练速度。这一模型具有良好的可扩展性,能适应大训练集的情况,是一种粗粒度的并行,且易于在集群系统这样的并行环境下进行数据挖掘。同时,描述了相关算法,并对训练速度进行了测试。
The paper analyzes the requirement of performance, mainly including validity, flexibility and parallelism of algorithm in data mining. Based on the idea of data parallelism, a parallel training model for RBF (radial basis function) neural network in time-series prediction to improve the training speed is proposed. This model has good expansibility can cope with large training set, and is a kind of coarse granular parallel and suitable to data mining on parallel system such as cluster. Also the algorithm is described and the training speed is tested.
出处
《计算机工程》
CAS
CSCD
北大核心
2005年第11期162-164,共3页
Computer Engineering
基金
校青年科学基金资助项目(2003Q14)
关键词
数据挖掘
数据预测
并行计算
神经网络
Data mining
Data prediction
Parallel computing
Neural network
