大规模数据的分布式神经网络回归模型研究  

Research on Distributed Neural Network Regression Model for Large-scale Data

作　　者：蔡超[1] 何馨怡 李丽[1] Cai Chao;He Xinyi;Li Li(School of Statistics,Shandong Technology and Business University,Yantai Shandong 264005,China)

机构地区：[1]山东工商学院统计学院,山东烟台264005

出　　处：《统计与决策》2023年第17期34-39,共6页Statistics & Decision

基　　金：国家社会科学基金资助项目(20BTJ052)。

摘　　要：随着计算机技术的飞速发展,大规模数据不断涌现,数据间呈现复杂的非线性特征,这使得传统的回归分析方法难以奏效。鉴于此,文章提出了基于交互有效方法的分布式神经网络回归(CE-RNN)模型,通过优化基于交互有效方法构建的神经网络回归模型的替代损失函数来获得全局参数估计值的近似结果。该模型一方面采用分布式计算方法避免了单台机器难以处理大规模数据的难题,另一方面使用神经网络回归模型解决了非线性回归问题。数值模拟和应用研究的结果表明:CE-RNN模型的预测性能与全局神经网络回归模型基本一致,且优于基于单轮型方法的分布式神经网络回归模型。With the rapid development of computer technology,large-scale data continue to emerge and show complex nonlinear characteristics,which makes the traditional regression analysis methods ineffective.On this basis,the paper proposes a distributed recurrent neural network regression model based on communication efficient approach(CE-RNN),which obtains approximate results of global parameter estimates by optimizing the surrogate loss function constructed based on the communication-efficient approach.On the one hand,the model adopts the distributed computing method to avoid the problem that a single machine is difficult to process large-scale data,and on the other hand,it uses the neural network regression model to solve the nonlinear regression problem.The results of numerical simulation and application research show that the prediction performance of the CE-RNN model is basically consistent with the global neural network regression model,and better than the distributed neural network regression model based on the One-shot method.

关 键 词：大规模数据 神经网络回归 分布式 交互有效 非线性 

分 类 号：O212.1[理学—概率论与数理统计]