Robust, Deep and Inductive Anomaly Detection

Raghavendra Chalapathy; Aditya Krishna Menon; Sanjay Chawla

doi:10.1007/978-3-319-71249-9_3

Robust, Deep and Inductive Anomaly Detection

Raghavendra Chalapathy, Aditya Krishna Menon^*, Sanjay Chawla

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

99 Citations (Scopus)

Abstract

PCA is a classical statistical technique whose simplicity and maturity has seen it find widespread use for anomaly detection. However, it is limited in this regard by being sensitive to gross perturbations of the input, and by seeking a linear subspace that captures normal behaviour. The first issue has been dealt with by robust PCA, a variant of PCA that explicitly allows for some data points to be arbitrarily corrupted; however, this does not resolve the second issue, and indeed introduces the new issue that one can no longer inductively find anomalies on a test set. This paper addresses both issues in a single model, the robust autoencoder. This method learns a nonlinear subspace that captures the majority of data points, while allowing for some data to have arbitrary corruption. The model is simple to train and leverages recent advances in the optimisation of deep neural networks. Experiments on a range of real-world datasets highlight the model’s effectiveness.

Original language	English
Title of host publication	Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings
Editors	Michelangelo Ceci, Saso Dzeroski, Celine Vens, Ljupco Todorovski, Jaakko Hollmen
Publisher	Springer Verlag
Pages	36-51
Number of pages	16
ISBN (Print)	9783319712482
DOIs	https://doi.org/10.1007/978-3-319-71249-9_3
Publication status	Published - 2017
Event	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017 - Skopje, Macedonia, The Former Yugoslav Republic of Duration: 18 Sept 2017 → 22 Sept 2017

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10534 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017
Country/Territory	Macedonia, The Former Yugoslav Republic of
City	Skopje
Period	18/09/17 → 22/09/17

Access to Document

10.1007/978-3-319-71249-9_3

Cite this

Chalapathy, R., Menon, A. K., & Chawla, S. (2017). Robust, Deep and Inductive Anomaly Detection. In M. Ceci, S. Dzeroski, C. Vens, L. Todorovski, & J. Hollmen (Eds.), Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings (pp. 36-51). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10534 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-319-71249-9_3

Chalapathy, Raghavendra ; Menon, Aditya Krishna ; Chawla, Sanjay. / Robust, Deep and Inductive Anomaly Detection. Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings. editor / Michelangelo Ceci ; Saso Dzeroski ; Celine Vens ; Ljupco Todorovski ; Jaakko Hollmen. Springer Verlag, 2017. pp. 36-51 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{28d9f85648de4d129512bcadfa0f8b30,

title = "Robust, Deep and Inductive Anomaly Detection",

abstract = "PCA is a classical statistical technique whose simplicity and maturity has seen it find widespread use for anomaly detection. However, it is limited in this regard by being sensitive to gross perturbations of the input, and by seeking a linear subspace that captures normal behaviour. The first issue has been dealt with by robust PCA, a variant of PCA that explicitly allows for some data points to be arbitrarily corrupted; however, this does not resolve the second issue, and indeed introduces the new issue that one can no longer inductively find anomalies on a test set. This paper addresses both issues in a single model, the robust autoencoder. This method learns a nonlinear subspace that captures the majority of data points, while allowing for some data to have arbitrary corruption. The model is simple to train and leverages recent advances in the optimisation of deep neural networks. Experiments on a range of real-world datasets highlight the model{\textquoteright}s effectiveness.",

keywords = "Anomaly detection, Autoencoders, Deep learning, Outlier detection, Robust PCA",

author = "Raghavendra Chalapathy and Menon, {Aditya Krishna} and Sanjay Chawla",

note = "Publisher Copyright: {\textcopyright} 2017, Springer International Publishing AG.; European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017 ; Conference date: 18-09-2017 Through 22-09-2017",

year = "2017",

doi = "10.1007/978-3-319-71249-9_3",

language = "English",

isbn = "9783319712482",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "36--51",

editor = "Michelangelo Ceci and Saso Dzeroski and Celine Vens and Ljupco Todorovski and Jaakko Hollmen",

booktitle = "Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings",

address = "Germany",

}

Chalapathy, R, Menon, AK & Chawla, S 2017, Robust, Deep and Inductive Anomaly Detection. in M Ceci, S Dzeroski, C Vens, L Todorovski & J Hollmen (eds), Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10534 LNAI, Springer Verlag, pp. 36-51, European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017, Skopje, Macedonia, The Former Yugoslav Republic of, 18/09/17. https://doi.org/10.1007/978-3-319-71249-9_3

Robust, Deep and Inductive Anomaly Detection. / Chalapathy, Raghavendra; Menon, Aditya Krishna; Chawla, Sanjay.
Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings. ed. / Michelangelo Ceci; Saso Dzeroski; Celine Vens; Ljupco Todorovski; Jaakko Hollmen. Springer Verlag, 2017. p. 36-51 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10534 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Robust, Deep and Inductive Anomaly Detection

AU - Chalapathy, Raghavendra

AU - Menon, Aditya Krishna

AU - Chawla, Sanjay

PY - 2017

Y1 - 2017

N2 - PCA is a classical statistical technique whose simplicity and maturity has seen it find widespread use for anomaly detection. However, it is limited in this regard by being sensitive to gross perturbations of the input, and by seeking a linear subspace that captures normal behaviour. The first issue has been dealt with by robust PCA, a variant of PCA that explicitly allows for some data points to be arbitrarily corrupted; however, this does not resolve the second issue, and indeed introduces the new issue that one can no longer inductively find anomalies on a test set. This paper addresses both issues in a single model, the robust autoencoder. This method learns a nonlinear subspace that captures the majority of data points, while allowing for some data to have arbitrary corruption. The model is simple to train and leverages recent advances in the optimisation of deep neural networks. Experiments on a range of real-world datasets highlight the model’s effectiveness.

AB - PCA is a classical statistical technique whose simplicity and maturity has seen it find widespread use for anomaly detection. However, it is limited in this regard by being sensitive to gross perturbations of the input, and by seeking a linear subspace that captures normal behaviour. The first issue has been dealt with by robust PCA, a variant of PCA that explicitly allows for some data points to be arbitrarily corrupted; however, this does not resolve the second issue, and indeed introduces the new issue that one can no longer inductively find anomalies on a test set. This paper addresses both issues in a single model, the robust autoencoder. This method learns a nonlinear subspace that captures the majority of data points, while allowing for some data to have arbitrary corruption. The model is simple to train and leverages recent advances in the optimisation of deep neural networks. Experiments on a range of real-world datasets highlight the model’s effectiveness.

KW - Anomaly detection

KW - Autoencoders

KW - Deep learning

KW - Outlier detection

KW - Robust PCA

UR - http://www.scopus.com/inward/record.url?scp=85040249118&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-71249-9_3

DO - 10.1007/978-3-319-71249-9_3

M3 - Conference contribution

SN - 9783319712482

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 36

EP - 51

BT - Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings

A2 - Ceci, Michelangelo

A2 - Dzeroski, Saso

A2 - Vens, Celine

A2 - Todorovski, Ljupco

A2 - Hollmen, Jaakko

PB - Springer Verlag

T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017

Y2 - 18 September 2017 through 22 September 2017

ER -

Chalapathy R, Menon AK, Chawla S. Robust, Deep and Inductive Anomaly Detection. In Ceci M, Dzeroski S, Vens C, Todorovski L, Hollmen J, editors, Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2017, Proceedings. Springer Verlag. 2017. p. 36-51. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-71249-9_3

Robust, Deep and Inductive Anomaly Detection

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this