A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification

Zihan Li; Wei Ding; Inal Mashukov; Scott Crouter; Ping Chen

doi:10.1007/978-981-97-2266-2_19

A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification

Zihan Li
, Wei Ding
, Inal Mashukov
, Scott Crouter
, Ping Chen

University of Massachusetts Boston
University of Tennessee, Knoxville

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

Abstract

Time series data plays a significant role in many research fields since it can record and disclose the dynamic trends of a phenomenon with a sequence of ordered data points. Time series data is dynamic, of variable length, and often contains complex patterns, which makes its analysis challenging especially when the amount of data is limited. In this paper, we propose a multi-view feature construction approach that can generate multiple feature sets of different resolutions from a single dataset and produce a fixed-length representation of variable-length time series data. Furthermore, we propose a multi-encoder-decoder Transformer (MEDT) architecture to effectively analyze these multi-view representations. Through extensive experiments using multiple benchmarks and a real-world dataset, our method shows significant improvement over the state-of-the-art methods.

Original language	English
Title of host publication	Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings
Editors	De-Nian Yang, Xing Xie, Vincent S. Tseng, Jian Pei, Jen-Wei Huang, Jerry Chun-Wei Lin
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	239-250
Number of pages	12
ISBN (Print)	9789819722655
DOIs	https://doi.org/10.1007/978-981-97-2266-2_19
State	Published - 2024
Event	28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024 - Taipei, Taiwan, Province of China Duration: May 7 2024 → May 10 2024

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14650 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024
Country/Territory	Taiwan, Province of China
City	Taipei
Period	5/7/24 → 5/10/24

ASJC Scopus Subject Areas

Theoretical Computer Science
General Computer Science

Keywords

Multi-Encoder-Decoder
Multi-view Learning
Multivariate Time Series Classification
Transformer
Variable-length Time Series

Access to Document

10.1007/978-981-97-2266-2_19

Cite this

Li, Z., Ding, W., Mashukov, I., Crouter, S., & Chen, P. (2024). A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification. In D.-N. Yang, X. Xie, V. S. Tseng, J. Pei, J.-W. Huang, & J. C.-W. Lin (Eds.), Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings (pp. 239-250). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14650 LNAI). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-97-2266-2_19

A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification. / Li, Zihan; Ding, Wei; Mashukov, Inal et al.
Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings. ed. / De-Nian Yang; Xing Xie; Vincent S. Tseng; Jian Pei; Jen-Wei Huang; Jerry Chun-Wei Lin. Springer Science and Business Media Deutschland GmbH, 2024. p. 239-250 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14650 LNAI).

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

Li, Z, Ding, W, Mashukov, I, Crouter, S & Chen, P 2024, A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification. in D-N Yang, X Xie, VS Tseng, J Pei, J-W Huang & JC-W Lin (eds), Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14650 LNAI, Springer Science and Business Media Deutschland GmbH, pp. 239-250, 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, Province of China, 5/7/24. https://doi.org/10.1007/978-981-97-2266-2_19

Li Z, Ding W, Mashukov I, Crouter S, Chen P. A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification. In Yang DN, Xie X, Tseng VS, Pei J, Huang JW, Lin JCW, editors, Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings. Springer Science and Business Media Deutschland GmbH. 2024. p. 239-250. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-981-97-2266-2_19

Li, Zihan ; Ding, Wei ; Mashukov, Inal et al. / A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification. Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings. editor / De-Nian Yang ; Xing Xie ; Vincent S. Tseng ; Jian Pei ; Jen-Wei Huang ; Jerry Chun-Wei Lin. Springer Science and Business Media Deutschland GmbH, 2024. pp. 239-250 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{be03260bb74849b2b4b515a235a2b76e,

title = "A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification",

abstract = "Time series data plays a significant role in many research fields since it can record and disclose the dynamic trends of a phenomenon with a sequence of ordered data points. Time series data is dynamic, of variable length, and often contains complex patterns, which makes its analysis challenging especially when the amount of data is limited. In this paper, we propose a multi-view feature construction approach that can generate multiple feature sets of different resolutions from a single dataset and produce a fixed-length representation of variable-length time series data. Furthermore, we propose a multi-encoder-decoder Transformer (MEDT) architecture to effectively analyze these multi-view representations. Through extensive experiments using multiple benchmarks and a real-world dataset, our method shows significant improvement over the state-of-the-art methods.",

keywords = "Multi-Encoder-Decoder, Multi-view Learning, Multivariate Time Series Classification, Transformer, Variable-length Time Series",

author = "Zihan Li and Wei Ding and Inal Mashukov and Scott Crouter and Ping Chen",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.; 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024 ; Conference date: 07-05-2024 Through 10-05-2024",

year = "2024",

doi = "10.1007/978-981-97-2266-2\_19",

language = "English",

isbn = "9789819722655",

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

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "239--250",

editor = "De-Nian Yang and Xing Xie and Tseng, \{Vincent S.\} and Jian Pei and Jen-Wei Huang and Lin, \{Jerry Chun-Wei\}",

booktitle = "Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7{\textendash}10, 2024, Proceedings",

}

TY - GEN

T1 - A Multi-view Feature Construction and Multi-Encoder-Decoder Transformer Architecture for Time Series Classification

AU - Li, Zihan

AU - Ding, Wei

AU - Mashukov, Inal

AU - Crouter, Scott

AU - Chen, Ping

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

PY - 2024

Y1 - 2024

N2 - Time series data plays a significant role in many research fields since it can record and disclose the dynamic trends of a phenomenon with a sequence of ordered data points. Time series data is dynamic, of variable length, and often contains complex patterns, which makes its analysis challenging especially when the amount of data is limited. In this paper, we propose a multi-view feature construction approach that can generate multiple feature sets of different resolutions from a single dataset and produce a fixed-length representation of variable-length time series data. Furthermore, we propose a multi-encoder-decoder Transformer (MEDT) architecture to effectively analyze these multi-view representations. Through extensive experiments using multiple benchmarks and a real-world dataset, our method shows significant improvement over the state-of-the-art methods.

AB - Time series data plays a significant role in many research fields since it can record and disclose the dynamic trends of a phenomenon with a sequence of ordered data points. Time series data is dynamic, of variable length, and often contains complex patterns, which makes its analysis challenging especially when the amount of data is limited. In this paper, we propose a multi-view feature construction approach that can generate multiple feature sets of different resolutions from a single dataset and produce a fixed-length representation of variable-length time series data. Furthermore, we propose a multi-encoder-decoder Transformer (MEDT) architecture to effectively analyze these multi-view representations. Through extensive experiments using multiple benchmarks and a real-world dataset, our method shows significant improvement over the state-of-the-art methods.

KW - Multi-Encoder-Decoder

KW - Multi-view Learning

KW - Multivariate Time Series Classification

KW - Transformer

KW - Variable-length Time Series

UR - https://www.scopus.com/pages/publications/85192842150

UR - https://www.scopus.com/pages/publications/85192842150#tab=citedBy

U2 - 10.1007/978-981-97-2266-2_19

DO - 10.1007/978-981-97-2266-2_19

M3 - Conference contribution

AN - SCOPUS:85192842150

SN - 9789819722655

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

SP - 239

EP - 250

BT - Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7–10, 2024, Proceedings

A2 - Yang, De-Nian

A2 - Xie, Xing

A2 - Tseng, Vincent S.

A2 - Pei, Jian

A2 - Huang, Jen-Wei

A2 - Lin, Jerry Chun-Wei

PB - Springer Science and Business Media Deutschland GmbH

T2 - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024

Y2 - 7 May 2024 through 10 May 2024

ER -