摘要
针对高维数据无监督异常检测难以重构异常样本,无法保留低维空间信息的问题,提出一种深度变分自编码高斯混合模型(deep variational autoencoding gaussian mixture model,DVAGMM)。该模型利用深度变分自编码器为每个输入样本生成低维数据和重构误差,并将这些数据输入高斯混合模型。为更好地学习到原始样本的低维特征,同时避免自编码器自身的局部优化问题,减少重构误差,模型采用联合优化深度变分自编码器和高斯混合模型参数的方法,并利用单独的估计网络促进混合模型的参数学习。实验结果表明,该模型在几个基准数据集上的检测准确率和效果都比其他传统模型更高,以F1值作为综合评价指标,模型的综合分数比第二名高出大约4%。
To address the problem that anomaly samples are difficult to reconstruct and low-dimensional spatial information unable to retain for unsupervised anomaly detection of high-dimensional data,a deep variational autoencoding Gaussian mixture model(DVAGMM)is proposed.A deep variational autoencoder is used to generate low-dimensional data and reconstruction errors for each input sample,and feed them into the Gaussian mixture model.In order to better learn the low-dimensional features of the original samples and reduce the reconstruction error occurring in the local optimization of the self-encoder itself,the model jointly optimizes the parameters of the depth-variant self-encoder and the Gaussian mixture model,and uses a separate estimation network to facilitate the parameter learning of the mixture model.Experimental results show that the model has higher detection accuracy and effectiveness than other conventional models on several benchmark datasets,and the combined score of the model is about 4%higher than the second place using the F1 score as a comprehensive evaluation metric.
作者
江连吉
陈玉明
钟才明
曾高发
JIANG Lianji;CHEN Yumin;ZHONG Caiming;ZENG Gaofa(School of Computer and Information Engineering,Xiamen University of Technology,Xiamen 361024,China;School of Information Engineering,College of Science and Technology,Ningbo University,Ningbo 315212,China;Zhixiang Intelligent Technology,Xiamen 361024,China)
出处
《厦门理工学院学报》
2023年第5期49-57,共9页
Journal of Xiamen University of Technology
基金
国家自然科学基金项目(61976183)。
关键词
变分自编码器
高斯混合模型
无监督异常检测
深度学习
联合训练
variational autoencoding
Gaussian mixture model
unsupervised anomaly detection
deep learning
joint optimization
