基于元学习和数据增强优化小样本模型泛化性能研究
Research on Small-sample Model Generalization Performance Optimization Based on Meta-Learning and Data Enhancement
邓天翊 1张耕培1
作者信息
- 1. 长江大学 电子信息与电气工程学院,湖北 荆州 434023
- 折叠
摘要
针对小样本模型泛化性能不足的问题,引入元学习机制构建强泛化性的数据分析模型.使用BP神经网络建立数据分析模型,并使用模型无关元学习算法MAML对模型进行优化.结果显示,相比于传统模型(如支持向量机和高斯过程方法),文中所建立模型的泛化性能更好;针对MAML训练数据形式,引入数据增强方法增加训练数据数量,文中所建立模型的均方根误差、平均绝对百分比误差和决定系数分别为0.05、0.066和0.85,均优于其他预测模型.
Abstract
To address the issue of insufficient generalization performance in small sample models,a Meta-Learning mechanism is introduced to construct a data analysis model with strong generalization.A BP neural network is used to establish a data analysis model,and Model-Agnostic Meta-Learning Algorithm(MAML)is used to optimize the model.The results show that compared to traditional models such as Support Vector Machine and Gaussian process methods,the model established in this paper has better generalization performance.Regarding the training data format of MAML,a data augmentation method is introduced to increase the amount of training data.The root mean square error,average absolute percentage error,and deciding coefficient of the model established in the paper are 0.05,0.066,and 0.85,respectively,which are superior to other prediction models.
关键词
元学习/优化/小样本模型/泛化性/模型无关元学习算法Key words
Meta-Learning/optimization/small sample model/generalization/MAML引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据