基于数据自律分发的神经网络集成模型

Neural Network Ensemble Based on Data Autonomous Allocation

针对神经网络在求解大规模数据时表现出的计算能力不足的瓶颈问题,本文在对神经网络集成理论及其算法进行分析研究的基础上,结合自律分散系统提出一种新的基于数据自律分发实现的神经网络集成模型,设计了该模型下由Pull-MA和Push-MA实现的确保时序一致性的通信机制,并给出了用于实现网络中数据自律分发和结果自律收集评价的训练算法。实验结果表明,所构建模型和集成算法对大规模数据的处理能达到理想的训练效果,网络具有良好的泛化和分类能力。

To solve the computing bottleneck problem of neural network processing large-scale data set, based on analyzing the autonomous decentralized system, the neural network ensemble theory and its training algorithm, a neural network autonomous set（NNAS） model realized by autonomous dataset allocation is proposed. In order to enable the system timeline, a network model communication mechanism implemented by pull mobile agent（Pull-MA） and push mobile agent （Push-MA） is designed, and a NNAS training algorithm is given to accomplish allocating dataset and collecting result autonomously. Experimental result shows that the model and its training algorithm can acquire ideal training effect, and the ensemble model can obtain better generalization and classification ability.