Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract)

Zahra Atashgahi; Decebal Constantin Mocanu; Raymond N.J. Veldhuis; Mykola Pechenizkiy

Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract)

Zahra Atashgahi^*, Decebal Constantin Mocanu, Raymond N.J. Veldhuis, Mykola Pechenizkiy

^*Corresponding author for this work

Research output: Contribution to conference › Paper › peer-review

142 Downloads (Pure)

Abstract

Change-point detection (CPD) is recognized as being one of the most significant tasks in time series analysis. While offline CPD has been vastly investigated in the last few years, online CPD still suffers from major challenges, such as high dependency on the choice of hyperparameters and prior information about data. However, in most real-world applications, very little prior information about the data is available, and performing hyperparameter tuning might not be feasible. Our proposed method, Adaptive LSTM-Autoencoder Change-point Detection (ALACPD), aims to address these challenges by performing unsupervised online memory-free CPD and continuously adapting itself to the current behavior of the system

Original language	English
Publication status	Published - Sept 2021
Event	8th ACM Celebration of Women in Computing womENcourage - virtual (coordinated from Prague, Czech Republic), Prague, Czech Republic Duration: 22 Sept 2021 → 24 Sept 2021 https://womencourage.acm.org/2021/

Conference

Conference	8th ACM Celebration of Women in Computing womENcourage
Country/Territory	Czech Republic
City	Prague
Period	22/09/21 → 24/09/21
Internet address	https://womencourage.acm.org/2021/

Access to Document

Extended AbstractFinal published version, 428 KB
PosterFinal published version, 1.16 MB

Cite this

@conference{f6916d3bb2854b85a13f26741be98feb,

title = "Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract)",

abstract = "Change-point detection (CPD) is recognized as being one of the most significant tasks in time series analysis. While offline CPD has been vastly investigated in the last few years, online CPD still suffers from major challenges, such as high dependency on the choice of hyperparameters and prior information about data. However, in most real-world applications, very little prior information about the data is available, and performing hyperparameter tuning might not be feasible. Our proposed method, Adaptive LSTM-Autoencoder Change-point Detection (ALACPD), aims to address these challenges by performing unsupervised online memory-free CPD and continuously adapting itself to the current behavior of the system",

author = "Zahra Atashgahi and Mocanu, {Decebal Constantin} and Veldhuis, {Raymond N.J.} and Mykola Pechenizkiy",

year = "2021",

month = sep,

language = "English",

note = "8th ACM Celebration of Women in Computing womENcourage ; Conference date: 22-09-2021 Through 24-09-2021",

url = "https://womencourage.acm.org/2021/",

}

Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract). / Atashgahi, Zahra; Mocanu, Decebal Constantin; Veldhuis, Raymond N.J. et al.
2021. Paper presented at 8th ACM Celebration of Women in Computing womENcourage, Prague, Czech Republic.

Research output: Contribution to conference › Paper › peer-review

TY - CONF

T1 - Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract)

AU - Atashgahi, Zahra

AU - Mocanu, Decebal Constantin

AU - Veldhuis, Raymond N.J.

AU - Pechenizkiy, Mykola

PY - 2021/9

Y1 - 2021/9

N2 - Change-point detection (CPD) is recognized as being one of the most significant tasks in time series analysis. While offline CPD has been vastly investigated in the last few years, online CPD still suffers from major challenges, such as high dependency on the choice of hyperparameters and prior information about data. However, in most real-world applications, very little prior information about the data is available, and performing hyperparameter tuning might not be feasible. Our proposed method, Adaptive LSTM-Autoencoder Change-point Detection (ALACPD), aims to address these challenges by performing unsupervised online memory-free CPD and continuously adapting itself to the current behavior of the system

AB - Change-point detection (CPD) is recognized as being one of the most significant tasks in time series analysis. While offline CPD has been vastly investigated in the last few years, online CPD still suffers from major challenges, such as high dependency on the choice of hyperparameters and prior information about data. However, in most real-world applications, very little prior information about the data is available, and performing hyperparameter tuning might not be feasible. Our proposed method, Adaptive LSTM-Autoencoder Change-point Detection (ALACPD), aims to address these challenges by performing unsupervised online memory-free CPD and continuously adapting itself to the current behavior of the system

M3 - Paper

T2 - 8th ACM Celebration of Women in Computing womENcourage

Y2 - 22 September 2021 through 24 September 2021

ER -

Unsupervised Online Memory-free Change-point Detection using an Ensemble of LSTM-Autoencoder-based Neural Networks (Extended Abstract)

Abstract

Conference

Access to Document

Fingerprint

Cite this