Anomaly detection via adaptive greedy model

Anomaly detection via adaptive greedy model
复制标题

通过自适应贪婪模型进行异常检测

DOI:
10.1016/j.neucom.2018.09.080
复制
发表时间:
2019-02
期刊:
影响因子:
6
通讯作者:
Xiaowei Xu
Xiaowei Xu
中科院分区:
计算机科学2区
文献类型:
--
作者:
Dongdong Hou;Yang Cong;Gan Sun;Ji Liu;Xiaowei Xu

文献摘要

参考文献

被引文献

相似文献

异常检测是各种研究领域和应用领域的基本问题之一。与大多数基于稀疏表示的异常检测方法采用稀疏性的松弛项通过1范数,提出了一种基于1范数约束的自适应贪婪模型优化的无监督异常检测方法,该方法在理论上具有更高的准确性、鲁棒性和稀疏性.首先,在特征表示方面,通过层叠自编码器网络以无监督的方式学习一个简洁的特征空间。本文提出了一种基于范数约束的字典选择模型,从训练数据中选择最优的小子集来构造压缩字典,在提高精度的同时降低了计算负担。最后,对每个测试样本进行基于范数约束的稀疏重构,并根据重构分数判断异常。在模型优化方面,利用自适应的前向-后向贪婪模型,在理论保证的前提下对该非凸问题进行优化。我们所提出的方法进行评估与我们的真实的工业数据集和基准数据集,各种实验结果表明,我们所提出的方法是与传统的监督方法相媲美,比大多数比较无监督的方法。
Anomaly detection is one of the fundamental problems within diverse research areas and application domains. In comparison with most sparse representation based anomaly detection methods adopting a relaxation term of sparsity via ℓ1norm, we propose an unsupervised anomaly detection method optimized via an adaptive greedy model based on ℓ0norm constraint, which is more accurate, robust and sparse in theory. Firstly for feature representation, a concise feature space is learned in an unsupervised way via stacked autoencoder network. We propose a dictionary selection model based on ℓ2, 0norm constraint to select an optimal small subset of the training data to construct a condense dictionary, which can improve accuracy and reduce computational burden simultaneously. Finally, each testing sample is reconstructed by ℓ0norm constraint based sparse representation, and anomalies are determined depending on the sparse reconstruction scores accordingly. For model optimization, an adaptive forward-backward greedy model is utilized to optimize this nonconvex problem with the theoretical guarantee. Our proposed method is evaluated with our real industrial dataset and benchmark datasets, and various experimental results demonstrate that our proposed method is comparable with conventional supervised methods and performs better than most comparative unsupervised methods.
DOI: 10.1016/j.neucom.2014.08.070
发表时间: 2015-02-03
期刊: NEUROCOMPUTING
影响因子: 6
作者:
Karami, Amin;Guerrero-Zapata, Manel
通讯作者: Guerrero-Zapata, Manel
DOI: 10.1109/tpami.2012.277
发表时间: 2013-08-01
影响因子: 23.6
作者:
Shin, Hoo-Chang;Orton, Matthew R.;Leach, Martin O.
通讯作者: Leach, Martin O.
DOI: 10.1109/tgrs.2014.2326654
发表时间: 2015-02-01
影响因子: 8.2
作者:
Yuan, Yuan;Wang, Qi;Zhu, Guokang
通讯作者: Zhu, Guokang
DOI: 10.1109/surv.2013.030713.00201
发表时间: 2013-01-01
影响因子: 35.6
作者:
Nadeem, Adnan;Howarth, Michael P.
通讯作者: Howarth, Michael P.
DOI: 10.1109/tip.2013.2260168
发表时间: 2013-04
影响因子: 10.6
作者:
Yang Cong;Ji Liu;Junsong Yuan;Jiebo Luo
通讯作者: Yang Cong;Ji Liu;Junsong Yuan;Jiebo Luo