Autoencoders for Time-Series with Hyperparameter Tuning

版本 1.1.2 (2.1 MB) 作者: Anika Terbuch
This toolbox enables the hyperparameter optimization using a genetic algorithm for autoencoders applied to multivariate time-series.
0.0
(0)
160.0 次下载
更新时间 2024/4/10
在 GitHub 上查看许可证
This toolbox enables the hyperparameter optimization using a genetic algoritm created with the toolbox "Generic Deep Autoencoder for Time-Series" which is also included in this framework.
The primary focus is on the hyperparameter optimization for autoencoders used for multi-channel time-series analysis using a meta-heuristic.
Each optimization is performed for a defined number of generations and for each hyperparameter setting three autoencoders are trained to get a better estimate of the performance of the architecutre on different folds of the data.
For more information on the autoencoder-architecture itself refer to https://de.mathworks.com/matlabcentral/fileexchange/111110-generic-deep-autoencoder-for-time-series
For the hyperparmaeter optimizaton a genetic algorithm combining two crossover operator for a better exploration of the serach space is used.
It is adviced to run this optimization on dedicated ML-training infrastructure, or other comparable infrastructure, to reduce the runtime.
Using this framework the following hyperparameters can be optimized:
- Beta-KL-Divergence (if beta-VAEs are used)
- LearningRate
- MiniBatchSize
- NeuronsDecoder (also for multiple layers)
- NeuronsEncoder (also for multiple layers)
- NumberEpochs
---------------------------------------------------------------------
This code can also be considered as supplemental Material to the Paper:
"Detecting Anomalous Multivariate Time-Series via Hybrid Machine Learning"
by Anika Terbuch, Paul O'Leary, Negin Khalili-Motlagh-Kasmaei, Peter Auer, Alexander Zöhrer and Vincent Winter
published in January 2023
This article investigates the use of hybrid machine
learning (HML) for the detection of anomalous multivariate timeseries (MVTS).
Focusing on a specific industrial use-case from
geotechnical engineering, where hundreds 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 50 physics-motivated key performance
indicators (KPIs) and 2) an unsupervised variational
autoencoder (VAE) with long short-term memory layers. The
KPI captures expert knowledge on the properties of the data
that infer the quality of produced elements; these are used as a
type of auto-labeling. 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
nonanomalous. The article compares the performance of different
VAE architectures [e.g., long short-term memory (LSTM-VAE)
and bidirectional LSTM (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 subprocesses yields
a better discriminating power and permits the identification
of interdependencies between the subprocesses. Interestingly,
different autoencoder architectures may be optimal for different
subprocesses; 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.
cite as:
@article{Terbuch2023Jan,
author = {Terbuch, Anika and O{'}Leary, Paul and Khalili-Motlagh-Kasmaei, Negin and Auer, Peter and Z{\ifmmode\ddot{o}\else\"{o}\fi}hrer, Alexander and Winter, Vincent},
title = {{Detecting Anomalous Multivariate Time-Series via Hybrid Machine Learning}},
journal = {IEEE Transactions on Instrumentation and Measurement},
volume = {72},
pages = {1--11},
year = {2023},
month = jan,
urldate = {2023-04-13},
issn = {1557-9662},
publisher = {IEEE},
doi = {10.1109/TIM.2023.3236354}
}
----------------------------------------------------------------------
More information on the functionality of the toolbox and the description of the framework can be found in the documents
1) Premeable_GA: Short general introduction to genetic algorithms and hyperparmaeter optimization
2) Intro_GA: Gives an introduction to the functionalities of the toolbox as well as algorithmic details and the syntax used
3) Example_HyperparameterOptimization: Provides an example of the workflow to perform hyperparameter optimization on a dataset available from Matlab 2022a on.

引用格式

Anika Terbuch (2024). Autoencoders for Time-Series with Hyperparameter Tuning (https://github.com/anikaTerbuch/Matlab-HPO_AE/releases/tag/1.1.2), GitHub. 检索时间: .

Terbuch, Anika, et al. “Detecting Anomalous Multivariate Time-Series via Hybrid Machine Learning.” IEEE Transactions on Instrumentation and Measurement, vol. 72, Institute of Electrical and Electronics Engineers (IEEE), 2023, pp. 1–11, doi:10.1109/tim.2023.3236354.

查看更多格式
MATLAB 版本兼容性
创建方式 R2022b
兼容任何版本
平台兼容性
Windows macOS Linux

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

AutoencoderDeep/mcode/@AutoencoderDeep

AutoencoderDeep/mcode/@HyperparametersAED

AutoencoderDeep/mcode/wrapper_functions

HyperparameterOptimization/mcode_GA

HyperparameterOptimization/mcode_GA/@HPOsettingsAED

HyperparameterOptimization/mcode_GA/supportFunctionsGA

AutoencoderDeep

AutoencoderDeep/Example

HyperparameterOptimization

版本 已发布 发行说明
1.1.2

See release notes for this release on GitHub: https://github.com/anikaTerbuch/Matlab-HPO_AE/releases/tag/1.1.2
1.1.1

Updated example
1.1.0

Title was changed
1.0.0

要查看或报告此来自 GitHub 的附加功能中的问题，请访问其 GitHub 仓库。
要查看或报告此来自 GitHub 的附加功能中的问题，请访问其 GitHub 仓库。