TARNet: Task-Aware Reconstruction for Time-Series Transformer
Open accessThe DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.
Abstract
Time-series data contains temporal order information that can guide representation learning for predictive end tasks (e.g., classification, regression). Recently, there are some attempts to leverage such order information to first pre-train time-series models by reconstructing time-series values of randomly masked time segments, followed by an end-task fine-tuning on the same dataset, demonstrating improved end-task performance. However, this learning paradigm decouples data reconstruction from the end task. We argue that the representations learnt in this way are not informed by the end task and may, therefore, be sub-optimal for the end-task performance. In fact, the importance of different timestamps can vary significantly in different end tasks. We believe that representations learnt by reconstructing important timestamps would be a better strategy for improving end-task performance. In this work, we propose TARNet, Task-Aware Reconstruction Network, a new model using Transformers to learn task-aware data reconstruction that augments end-task performance. Specifically, we design a data-driven masking strategy that uses self-attention score distribution from end-task training to sample timestamps deemed important by the end task. Then, we mask out data at those timestamps and reconstruct them, thereby making the reconstruction task-aware. This reconstruction task is trained alternately with the end task at every epoch, sharing parameters in a single model, allowing the representation learnt through reconstruction to improve end-task performance. Extensive experiments on tens of classification and regression datasets show that TARNet significantly outperforms state-of-the-art baseline models across all evaluation metrics.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 1×
“Learning data representation through reconstruction has been explored in natural language processing [7] and time-series [37].”
BackgroundXGBoost: A Scalable Tree Boosting System
Chen, T., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939785 · Cited by 54,569 · cited 1×
“(3) Tree-based methods: Random Forest [26] and XGBoost [4].”
MethodSupport Vector Regression Machines
Drucker, H., Burges, C. J. C., Kaufman, L., Smola, A. J., & Vapnik, V. N. (1996) · Cited by 4,658
Random Forest: A Classification and Regression Tool for Compound Classification and QSAR Modeling
Svetnik, V., Liaw, A., Tong, C., Culberson, J. C., Sheridan, R. P., & Feuston, B. P. (2003) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci034160g · Cited by 3,573 · cited 1×
“(3) Tree-based methods: Random Forest [26] and XGBoost [4].”
MethodA Transformer-based Framework for Multivariate Time Series Representation Learning
Zerveas, G., Jayaraman, S., Patel, D., Bhamidipaty, A., & Eickhoff, C. (2021) · Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining · doi:10.1145/3447548.3467401 · Cited by 1,278 · cited 10×
“Learning data representation through reconstruction has been explored in natural language processing [7] and time-series [37].”
BackgroundMethodResultInfluentialMultivariate LSTM-FCNs for time series classification
Karim, F., Majumdar, S., Darabi, H., & Harford, S. (2019) · Neural Networks · doi:10.1016/j.neunet.2019.04.014 · Cited by 1,138 · cited 2×
“(2) MLSTM-FCNs [18] expands LSTM-FCN and Attention LSTMFCN by adding squeeze-and-excitation blocks.”
BackgroundMethodDempster, A., Petitjean, F., & Webb, G. I. (2020) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-020-00701-z · Cited by 982 · cited 3×
“TARNet performs the best on 17 datasets, as compared to 7 and 6 by the next best baselines TST [37] and Rocket [5], respectively.”
BackgroundMethodTime-Series Representation Learning via Temporal and Contextual ContrastingPreprint
Eldele, E., Ragab, M., Chen, Z., Wu, M., Kwoh, C., Li, X.-L., & Guan, C.-T. (2021) · International Joint Conference on Artificial Intelligence · doi:10.48550/arxiv.2106.14112 · Cited by 968 · cited 2×
“(9) TS-TCC [11] encourages consistency of different data augmentations to learn transformation-invariant representations.”
BackgroundTS2Vec: Towards Universal Representation of Time Series
Yue, Z., Wang, Y., Duan, J., Yang, T., Huang, C., Tong, Y., & Xu, B. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i8.20881 · Cited by 850 · cited 1×
“(7) TS2Vec [35] performs hierarchical contrastive learning over augmented context views.”
BackgroundTime Series Classification Using Multi-Channels Deep Convolutional Neural Networks
Zheng, Y., Liu, Q., Chen, E., Ge, Y., & Zhao, J. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-08010-9_33 · Cited by 652 · cited 1×
“Most neural networksbased methods use some arrangement of LSTM, CNN or both [18, 39].”
MethodFast time series classification using numerosity reduction
Xi, X., Keogh, E., Shelton, C., Wei, L., & Ratanamahatana, C. A. (2006) · Proceedings of the 23rd international conference on Machine learning - ICML '06 · doi:10.1145/1143844.1143974 · Cited by 591 · cited 1×
“Distance-based methods [8, 31] use distance metric to measure similarity of a pair of time series.”
MethodMiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series Classification
Dempster, A., Schmidt, D. F., & Webb, G. I. (2021) · Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining · doi:10.1145/3447548.3467231 · Cited by 490 · cited 2×
“(6) MiniRocket [6] upgrades Rocket by speeding it up, using a small, fixed set of kernels, and is almost entirely deterministic.”
BackgroundUnsupervised Scalable Representation Learning for Multivariate Time Series
Franceschi, J.-Y., Dieuleveut, A., & Jaggi, M. (2019) · arXiv · doi:10.48550/arxiv.1901.10738 · Cited by 305 · cited 2×
“Unsupervised representation learning for time-series uses triplet loss with negative sampling [14], hierarchical contrastive loss [36], temporal and contextual contrasting [11], local smoothness to define neighborhoods in time [28], and reprogramming acoustic models [32].”
BackgroundUnsupervised representation learning for time series with temporal neighborhood coding
Tonekaboni, S., Eytan, D., & Goldenberg, A. (2021) · arXiv · doi:10.48550/arxiv.2106.00750 · Cited by 170 · cited 2×
“Unsupervised representation learning for time-series uses triplet loss with negative sampling [14], hierarchical contrastive loss [36], temporal and contextual contrasting [11], local smoothness to define neighborhoods in time [28], and reprogramming acoustic models [32].”
BackgroundLearning a symbolic representation for multivariate time series classification
Baydogan, M. G., & Runger, G. (2015) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-014-0349-y · Cited by 169 · cited 1×
“[3] introduced Symbolic Representation to learn local relationships between different dimensions.”
BackgroundShapeNet: A Shapelet-Neural Network Approach for Multivariate Time Series Classification
Li, G., Choi, B., Xu, J., S Bhowmick, S., Chun, K.-P., & Wong, G. L.-H. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i9.17018 · Cited by 169 · cited 2×
“(5) ShapeNet [20] extends shapelet [33] for multivariate timeseries.”
BackgroundOn the Non-Trivial Generalization of Dynamic Time Warping to the Multi-Dimensional Case
Shokoohi-Yekta, M., Wang, J., & Keogh, E. (2015) · Proceedings of the 2015 SIAM International Conference on Data Mining · doi:10.1137/1.9781611974010.33 · Cited by 144 · cited 1×
“Euclidean Distance (ED), dimension independent dynamic timewarping (DTWI), and dimensiondependent dynamic time warping (DTWD) [25].”
BackgroundMultivariate Time Series Classification with WEASEL+MUSE
Schäfer, P., & Leser, U. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.11343 · Cited by 108 · cited 2×
“WEASEL-MUSE [24] utilizes bag of SFA (Symbolic Fourier Approximation).”
MethodLiang, H., Song, L., Wang, J., Guo, L., Li, X., & Liang, J. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.10.084 · Cited by 106 · cited 1×
“Depending on the end task, certain timestamps in time-series data may be more important than others [21], which the learnt representation ignores.”
BackgroundHsieh, T.-Y., Wang, S., Sun, Y., & Honavar, V. (2021) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3437963.3441815 · Cited by 96
The UEA multivariate time series classification archive, 2018
Bagnall, A., Dau, H. A., Lines, J., Flynn, M., Large, J., Bostrom, A., Southam, P., & Keogh, E. (2018) · arXiv · doi:10.48550/arxiv.1811.00075 · Cited by 87
Efficient Learning of Timeseries Shapelets
Hou, L., Kwok, J., & Zurada, J. (2016) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v30i1.10178 · Cited by 75 · cited 1×
“dynamic shapelets [23], efficient shapelets [16].”
MuVAN: A Multi-view Attention Network for Multivariate Temporal Data
Yuan, Y., Xun, G., Ma, F., Wang, Y., Du, N., Jia, K., Su, L., & Zhang, A. (2018) · 2018 IEEE International Conference on Data Mining (ICDM) · doi:10.1109/icdm.2018.00087 · Cited by 66 · cited 1×
“We also highlight that the utility of self-attention goes beyond computing internal data representation of a model to improve performance [29] or providing meaningful explanations [17, 34].”
BackgroundTowards Automatic Spatial Verification of Sensor Placement in Buildings
Hong, D., Ortiz, J., Whitehouse, K., & Culler, D. (2013) · ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys) · doi:10.1145/2528282.2528302 · Cited by 39
Self-citationTS2Vec: Towards Universal Representation of Time SeriesPreprint
Yue, Z., Wang, Y., Duan, J., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.10466 · Cited by 32 · cited 1×
“(7) TS2Vec [35] performs hierarchical contrastive learning over augmented context views.”
BackgroundAdversarial Dynamic Shapelet Networks
Ma, Q., Zhuang, W., Li, S., Huang, D., & Cottrell, G. W. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i04.5948 · Cited by 29 · cited 1×
“dynamic shapelets [23], efficient shapelets [16].”
Correlative Channel-Aware Fusion for Multi-View Time Series Classification
Bai, Y., Wang, L., Tao, Z., Li, S., & Fu, Y. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i8.16830 · Cited by 28 · cited 2×
“, learnable temporal pooling [19], correlative channel-aware learnable fusion [2], label-learning [22], attentional prototype network [38], and shapelet embedding [20].”
BackgroundJoint-Label Learning by Dual Augmentation for Time Series Classification
Ma, Q., Zheng, Z., Zheng, J., Li, S., Zhuang, W., & Cottrell, G. W. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i10.17071 · Cited by 19 · cited 1×
“, learnable temporal pooling [19], correlative channel-aware learnable fusion [2], label-learning [22], attentional prototype network [38], and shapelet embedding [20].”
BackgroundLearning Discriminative Virtual Sequences for Time Series Classification
Dorle, A., Li, F., Song, W., & Li, S. (2020) · Proceedings of the 29th ACM International Conference on Information & Knowledge Management · doi:10.1145/3340531.3412099 · Cited by 11 · cited 1×
“Distance-based methods [8, 31] use distance metric to measure similarity of a pair of time series.”
MethodLearnable Dynamic Temporal Pooling for Time Series ClassificationPreprint
Lee, D., Lee, S., & Yu, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.02577 · Cited by 1 · cited 1×
“, learnable temporal pooling [19], correlative channel-aware learnable fusion [2], label-learning [22], attentional prototype network [38], and shapelet embedding [20].”
Background
Research works printed but not linked: 9
- Hassan Ismail Fawaz, Germain Forestier, Jonathan Weber, Lhassane Idoumghar, and Pierre-Alain Muller. 2019. Deep learning for time series classification: a review. Data mining and knowledge discovery 33, 4 (2019), 917-963.
- Hassan Ismail Fawaz, Benjamin Lucas, Germain Forestier, Charlotte Pelletier, Daniel F Schmidt, Jonathan Weber, Geoffrey I Webb, Lhassane Idoumghar, Pierre- Alain Muller, and François Petitjean. 2020. Inceptiontime: Finding alexnet for time series classification. Data Mining and Knowledge Discovery 34, 6 (2020).
- Chang Wei Tan, Christoph Bergmeir, Francois Petitjean, and Geoffrey I Webb. 2020. Monash university, uea, ucr time series regression archive. arXiv e-prints (2020), arXiv-2006.
- Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems. 5998-6008.
- Zhiguang Wang, Weizhong Yan, and Tim Oates. 2017. Time series classification from scratch with deep neural networks: A strong baseline. In 2017 International joint conference on neural networks (IJCNN). IEEE, 1578-1585.
- Chao-Han Huck Yang, Yun-Yun Tsai, and Pin-Yu Chen. 2021. Voice2series: Re- programming acoustic models for time series classification. In ICML.
- Lexiang Ye and Eamonn Keogh. 2009. Time series shapelets: a new primitive for data mining. In KDD. 947-956.
- Zhihan Yue, Yujing Wang, Juanyong Duan, Tianmeng Yang, Congrui Huang, and Bixiong Xu. 2021. Learning Timestamp-Level Representations for Time Series with Hierarchical Contrastive Loss. arXiv preprint arXiv:2106.10466 (2021). · arXiv
- Xuchao Zhang, Yifeng Gao, Jessica Lin, and Chang-Tien Lu. 2020. Tapnet: Multi- variate time series classification with attentional prototypical network. In AAAI.
Cited by
Deep Learning for Time Series Classification and Extrinsic Regression: A Current Survey
Mohammadi Foumani, N., Miller, L., Tan, C. W., Webb, G. I., Forestier, G., & Salehi, M. (2024) · ACM Computing Surveys · doi:10.1145/3649448 · Cited by 268
“…2021 CNN + Transformers Sequence masking Voice2Series [22] 2021 CNN+Transformers Binary masking TST [19] 2021 Transformers Binary masking TARNet [150] 2022 Transformers Binary masking TimeMAE [151] 2023 CNN + Transformers Sequence masking CRT [152] 2023 Transformers Sequence masking Other…”
MethodSelf-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects
Zhang, K., Wen, Q., Zhang, C., Cai, R., Jin, M., Liu, Y., Zhang, J. Y., Liang, Y., Pang, G., Song, D., et al. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3387317 · Cited by 295
“Unlike random masking, TARNet [77] considers the pre-trained model based on the masking strategy irrelevant to the downstream task, which leads to sub-optimal representations.”
MethodA Survey on Time-Series Pre-Trained Models
Ma, Q., Liu, Z., Zheng, Z., Huang, Z., Zhu, S., Yu, Z., & Kwok, J. T.-Y. (2024) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2024.3475809 · Cited by 182
“Transformers have received much attention because of their ability to capture input data’s contextual (future-past) dependencies, making them a good choice for modeling sequence data [122].”
BackgroundGinAR: An End-To-End Multivariate Time Series Forecasting Model Suitable for Variable Missing
Yu, C., Wang, F., Shao, Z., Qian, T., Zhang, Z., Wei, W., & Xu, Y. (2024) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3637528.3672055 · Cited by 83
“On the one hand, since each missing variable is usually a straight-line sequence composed of outliers, the sequence model in STGNNs [16] cannot mine any valuable pattern and information, resulting in incorrect temporal dependencies.”
BackgroundYao, B., Li, X., Ji, Z., Xiao, K., & Xu, W. (2023) · Computers & Industrial Engineering · doi:10.1016/j.cie.2023.109855 · Cited by 63
Metadata
- Authors
- Chowdhury, R. R., Zhang, X., Shang, J., Gupta, R. K., & Hong, D.
- Affiliations
- University of California San Diego (US)
- Amazon (United States) (US)
- Venue
- Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
- Publisher
- ACM
- Published
- 2022-08-14
- Pages
- 212-220
- OpenAlex
- W4290877088
- Cited by
- 68
- References
- 21
- Citations by year
- 2026: 14 · 2025: 21 · 2024: 23 · 2023: 9 · 2022: 1
- FWCI
- 9.60
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) (publishedVersion, cc-by)
- Funders
- Semiconductor Research Corporation (2018-JU-2779)
- National Science Foundation Convergence Accelerator (OIA-2040727)
- Topics
- Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
- Social Sciences › Decision Sciences › Management Science and Operations Research › Stock Market Forecasting Methods
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Keywords
- time series representation learning, temporal masking, time series classification, time series regression
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.