Do deep neural networks contribute to multivariate time series anomaly detection?

Audibert, Julien; Michiardi, Pietro; Guyard, Frédéric; Marti, Sébastien; Zuluaga, María A.

doi:10.1016/j.patcog.2022.108945

Cited by 49 publications

(50 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…USAD 30 : This method performs adversarial training inspired by generative algorithms, enabling it to learn how to amplify the reconstruction error of anomalous inputs.…”

Section: Methodsmentioning

confidence: 99%

Graph‐based Bayesian network conditional normalizing flows for multiple time series anomaly detection

Xie

Ning

Huang

et al. 2022

Int J of Intelligent Sys

View full text Add to dashboard Cite

Various devices and sensors of cyber‐physical systems interact with each other in time and space, and the generated multiple time series have implicit correlations and highly nonlinear relationships. Determining how to model the multiple time series and capture dependencies through extracting features is the key to anomaly detection. In this paper, we propose a graph‐based Bayesian network conditional normalizing flows model for multiple time series anomaly detection, Bayesian network conditional normalizing flows (BNCNF). It applies a Bayesian network to model the causal relationships of multiple time series and introduces a spectral temporal dependency encoder to obtain the representations of interdependency between multiple time series. The representations are introduced as conditional information into the normalizing flows for density estimation, and the data corresponding to low density is judged as anomalies. The experiment are conducted on SWaT, WADI, and SMD datasets, the F1 score reaches 0.95, 0.92, and 0.97 on the three datasets, respectively. The results show that BNCNF has better performance in anomaly detection compared with the current mainstream methods.

show abstract

“…USAD 30 : This method performs adversarial training inspired by generative algorithms, enabling it to learn how to amplify the reconstruction error of anomalous inputs.…”

Section: Methodsmentioning

confidence: 99%

Graph‐based Bayesian network conditional normalizing flows for multiple time series anomaly detection

Xie

Ning

Huang

et al. 2022

Int J of Intelligent Sys

View full text Add to dashboard Cite

show abstract

“…With the presence of anomalies in the training set, this threshold turns out to be too high, and thus prevents the models from detecting all the anomalies. Previous works such as [16], [24], tested different thresholds and reported those that led to the best F1-score which explains why they got better results. On the other hand, the performance of iForest model reaches the same performance level as OC-SVM but decreases afterwards with the increase of the contamination ratio.…”

Section: A Performance With the Mixed-datasetmentioning

confidence: 97%

“…They categorized different approaches based on the nature of the input data, the type of anomaly (referred to as outlier in the survey) and the type of method. Focusing on unsupervised learning approaches for multivariate time-series data, Audibert et al [16], proposed a different taxonomy by classifying the methods into conventional methods, machine-learning methods and deep-learning methods. The conventional methods, such as VAR (Vector Auto Regressive) model, PCA (Principal Components Analysis) or SSA (Singular Spectrum Analysis), rely on statistical rules or assume that the time-series data come from a (linear) model whose parameters are estimated.…”

Section: B Unsupervised Learning Models For Anomaly Detectionmentioning

confidence: 99%

Assessing Unsupervised Machine Learning solutions for Anomaly Detection in Cloud Gaming Sessions

Mathieu

Lahmadi

et al. 2022

2022 18th International Conference on Network and Service Management (CNSM)

View full text Add to dashboard Cite

Cloud gaming applications have gained great adoption on the Internet particularly benefiting from the wide availability of broadband access networks. However, they still fail to meet users' quality requirements when accessed using cellular networks due to common wireless channel degradations. Machine Learning (ML) techniques can be leveraged to detect such anomalies during users' cloud gaming sessions. In this respect, unsupervised ML approaches are particularly interesting since they do not require labeled datasets. In this work, we investigate these approaches to understand their performance and their robustness. Our dataset consists of game sessions played on the public Google Stadia Cloud Gaming servers. The game sessions are played using a 4G network emulation replicating the capacity variations sampled on a commercial 4G network. We compare different models ranging from traditional approaches to deep learning and we evaluate their default performance while varying the level of contamination in their training datasets. Our experiments show that Auto-Encoders models achieve the best performance without contamination while the OC-SVM and the Isolation Forest are the most robust to data contamination.

show abstract

“…Because of these circumstances, an unsupervised anomaly detection approach was chosen. Unsupervised anomaly detection is done under the assumption that the training set Γ ⊂ Φ, whereas Φ is the set of all available samples , contains mostly samples from normal operations but could contain a small number of still unknown anomalies [13], [40]. This training set is constructed using the outlierness calculated with the KPI [13] for this reason, we call it KPI-supervised training as the anomalous samples are excluded and autolabeled using the outcome of the KPI classifier.…”

Section: Kpi-supervised Trainingmentioning

confidence: 99%

“…Especially in anomaly detection, benchmarks need to be treated with care, since most suffer from one or all of the following issues [12]: the problem is too trivial and can be solved in a few lines of code; the anomaly density is too high; if supervised learning is used, the ground truth is mislabeled and often there is no data available after the anomaly has occurred, this the authors of [12] called the run to failure bias. The study done on the comparison of classical pure ML methods, pure (deep) machine learning methods, and the combination of both, hybrid machine learning, suggests that there is no pure machine learning method that can outperform a hybrid system [13].…”

Section: Introductionmentioning

confidence: 99%

Detecting Anomalous Multivariate Time-Series via Hybrid Machine Learning

Terbuch

O’Leary

Khalili‐Motlagh‐Kasmaei

et al. 2023

IEEE Trans. Instrum. Meas.

View full text Add to dashboard Cite

This paper investigates the use of hybrid machine learning (HML) for the detection of anomalous multivariate time-series (MVTS). Focusing on a specific industrial use case from geotechnical engineering, where 100's of MVTS need to be analyzed and classified, has permitted extensive testing of the proposed methods with real measurement data. The novel hybrid anomaly detector combines two means for detection, creating redundancy and reducing the risk of missing defective elements in a safety relevant application. The two parts are: (1) anomaly detection based on approximately fifty physics-motivated key performance indicators (KPI) and ( 2) an unsupervised variational autoencoder with long short-term memory layers. The KPI capture expert knowledge on properties of the data that infer the quality of produced elements; these are used as a type of autolabeling. The goal of the extension using machine learning (ML) is to detect anomalies that the experts may not have foreseen. In contrast to anomaly detection in streaming data, where the goal is to locate an anomaly, each MVTS is complete in itself at the time of evaluation and is categorized as anomalous or non-anomalous.The paper compares the performance of different variationalautoencoder architectures (e.g., LSTM-VAE and BiLSTM-VAE). The results of using a genetic algorithm to optimize the hyperparameters of the different architectures are also presented. It is shown that modeling the industrial process as an assemblage of sub-processes yields a better discriminating power and permits the identification of inter-dependencies between the sub-processes. Interestingly, different autoencoder architectures may be optimal for different sub-processes; here two different architectures are combined to achieve superior performance. Extensive results are presented based on a very large set of real-time measurement data.

show abstract

Do deep neural networks contribute to multivariate time series anomaly detection?

Cited by 49 publications

References 21 publications

Graph‐based Bayesian network conditional normalizing flows for multiple time series anomaly detection

Graph‐based Bayesian network conditional normalizing flows for multiple time series anomaly detection

Assessing Unsupervised Machine Learning solutions for Anomaly Detection in Cloud Gaming Sessions

Detecting Anomalous Multivariate Time-Series via Hybrid Machine Learning

Contact Info

Product

Resources

About